I know my students have been waiting for (and possibly dreading) this day.

When I took over the class back in February, I was so, so, naive. I knew I was going in blind, so the first day I walked into that classroom I was ready with two things: first, a list of questions/ice breakers and second, a written assessment for each student to complete.

No ice was broken that first day. High School kids don’t want to speak out in a group, nor do they even want to raise their hands for a poll. While I’d soon make progress on both of these fronts, on that first day I faced a bunch of teenagers staring at me blankly (or worse, at their phone under the table).

One question got their attention:

“Who’s planning on taking the AP test?”

Around half the class shot their hands up, followed by an unsolicited outcry of complaints about how they weren’t ready, they had learned nothing, they weren’t going to take it just to fail, choice comments about the prior teacher, etc.

Sign up for an AP class, you take the test. Do it for college credit, but more importantly, it strengthens your app, showing colleges you’re motivated by and capable of college level curriculum.

The second thing I brought — the assessment — was mainly to gauge what they had learned so far, but at the end I threw in a set of qualitative questions like “What are you worried about?” and “What’s your confidence level from 1-5.”

Every single kid was worried about the test and just one kid — who it turned out had taken some JC programming courses already — put their confidence at a three. Most kids put down two; a lot put down one, gotta few zeros.

I collected the assessments and looked them over before I left. I told the class that IF I signed up to be their teacher — every single one of them would be confident walking into that test; I’d make sure they were ready.

In my head I was already figuring out how to get them back on track; of course I took on the class.

I structured everything I taught with the shared, singular goal of passing the test. To keep them focused, I kept a running countdown on the whiteboard: days left before the AP test.

I’d never taught High School before, never mind an AP course with its rigidly defined curriculum. I tried to learn it all. What were the critical elements of the course, what were the deadlines, how did grading work, what tools were available to me?

Covering a year’s worth of material in three months was a challenge I was up for, but my teenagers had other plans.

I’ve taught loads of adults going way back to when I was in my 20s, freelance teaching adult school PC classes.

But High School is different.

High School carries a romance wrapped up in memories, propped up by yearbooks and colored in by teen movies full of lovable nerds, mockable jocks, mean girls, principals doing the right thing and teachers who are always there for the kids.

It’s not quite like that.

I quickly learned I couldn’t talk - heck I couldn’t do any one thing (unless it was a test and then they begged for more time) — for more than 10 minutes. Longer than 10 minutes and I’d lose them.

There I am — up there at the board getting excited, waving my hands and telling stories, going through concepts and problems looking back to see if they’re getting it, when some kid raises his hand.

Full eye contact.

I get excited.

“Yes, you have a question?”

“Can I go to the bathroom?”

And I fell for it — Every. Single. Time.

As they got comfortable with me, some kids would occasionally ask a real question. But they were often poorly articulated and by the time I figured out what they were asking, I’d lost half the class.

Wednesdays and Fridays were 90 minute marathons. I needed to structure my lesson plan as 9 interconnected mini-classes. I got better at this as I went: lecture, video, hands-on coding, classwork (gotta be graded or else they won’t do it), group work, Kahoot! quizzes (with prizes), human stories, visualizations, etc.

High school kids have a lot going on. I give them total grace here. I don’t know what’s up with their home life, the high school drama they’re dealing with, or how much sleep they got the night before. They each juggle at least six classes apiece across a schedule I never fully figured out. I had a couple kids on the baseball team and they’d regularly miss class or have to leave early. Most days a paper would arrive from the office pulling some kid out for an appointment or club.

Perhaps that’s why they were so awful at due dates and assignments?

I’d walk the room while they were coding and most of the kids talked to me and showing me what they were working on and we’d talk about it. Then there were the quiet ones who didn’t ask for help.

I made a mistake trusting the quiet ones.

AP CSP is an outlier in that you have to submit a coding project a month before the written exam. The project is worth 30% of your score. The class was behind, so my initial focus was getting that coding project done and submitted. My first month with the class was all about the project and once every kid had completed it, I moved on.

I had three more units to cover.

Then, the official AP submission date for the coding project snuck up. One day, the school’s AP administrator showed up in my class with a list of kids who hadn’t submitted their project online yet and it was 1/3 of the class!

When we were at T-minus a few days, that list was down to 3 kids. All the quiet ones.

I gave the rest of the class a programming assignment, and I pulled those three kids out and sat them together at the table right in front of me. They did not appreciate the attention. One kid was just being lazy and hadn’t gone through the submission steps. One kid was nearly done and just needed a little push.

But the last kid was spiraling. He had created an elaborate choose-your-adventure game with nearly 300 lines of code. For context, the cleanest submissions were at most 50 lines of code.

This poor kid was staying up late every night working on it but he just kept thinking of new scenarios and adding them with no end in sight. And it was full of bugs. He was freaked out. He freaked me out. We debugged it together in class, reviewed the rubric, talked about KISS (Keep it Simple …), and I made him pinky promise to prune it down and just focus on getting it in. The school had a random off day coming up, so I wouldn’t see him again before it was due.

The bell rang. Watching him walk out the door, I had zero confidence he would get it in on time. Without the coding project, there’s just no way you’re passing the test. I checked online the night before it was due — he still hadn’t submitted it.

But when I checked the next morning, he had gotten it in just under the wire with only minutes to spare. Fistbump the next time i saw him.

On my last day with the class, I pulled the AP curriculum poster down off the whiteboard where I’d pinned it and they all signed it for me.

He wrote:

Another assumption I made — good High School, small AP class equals motivated kids. Incorrect.

At first I was determined to bring every kid along, but gradually I realized some just wouldn’t do the work. One kid — and he was smart just like every other kid in that room — never turned in an assignment.

He showed up early one day and I had a talk with him and he promised to turn stuff in.

He did not turn anything in.

And it’s not like they were malicious. Every single kid in that class — even the most annoying ones that I yelled at, or continually tried to play me — they were all good natured. I’ll never know why that kid didn’t do the work, because I know he was capable and I know he was curious.

The day after the test I brought in boxes of donuts. As they stuffed their faces we did a little test de-brief.

How’d it go?

Every kid told me something similar: I was clutch and the GOAT and I saved them and that they passed the test. Even that kid that didn’t do any work was supremely confident. And I want to believe it — they all showed a remarkable ability to synthesize and learn the material and if the homework and class tests weren’t motivation, the AP Test sure was.

So how’d my kids actually do? I don’t know.

Getting an account to access the college board website as a sub was hard. Getting an account to review my kids project submissions was stupid hard.1 Getting an account to see my test scores is impossible and it’s summer and my schools AP coordinator is out.

While I don’t know how my kids did, instead, thanks to the College Board, I know how everyone else’s kids did:

163,000 kids is around a 7% drop from last year, which is the first significant drop in this course in nearly a decade. My anecdotal conclusion is kids are correlating the rise of AI and the doomer headlines and concluding a career in Computer Science is not where it’s at.

As an academic discipline, Computer Science is a young 60 years old versus subjects like Physics (400 years), Calculus (350), and Chemistry (250). While all of those are pretty static, AI shows us that CS is far from done.

I’m still surprised how my little hobby has blown up across nearly every aspect of society and with AI, the rate of change is increasing. A lot of what I taught these kids for the test wasn’t so different from what I learned in my programming classes 40 years ago.

To make it real, we talked about AI — a lot. They had questions: was it using all the water, would programmers still have jobs, what even was ChatGPT?

AI is excluded from the AP curriculum. You’re not allowed to use it for coursework or coding.

So when I got this email announcing a course revamp around AI I was excited:

But this is two years away. With 17 million high school kids in the US, this stuff is slow to change. Who knows what AI will look like in two years.

My school isn’t offering any Computer Science classes next year. But it doesn’t take an entire year to understand how AI works — what it’s good for, what it’s not. It’s not magic.

Eventually, the school’s AP administrator will come back to the office and I’ll find out how my class did, but I do know how one of my best students did.

On Monday, the day the test scores came out, I got an email from her. She was excited to report she got a FIVE!

Let’s call her June and she’s all kinds of awesome. June’s coding project was a build-a-burger ASCII art app which inspired my AI demo app to introduce a post-test AI vibe coding unit.

On my last day of teaching, we just sat around and talked. Even though it was optional (not graded) most of the kids built something with AI. On that last day, a few kids were still building, so I was there to help, but mostly I let them do whatever they wanted. June was done, and she asked if they could play volleyball. There was a playground just outside my classroom so I said why not.

Too late to fire me.

A volleyball emerged from a backpack and June and a handful of my other kids had the best time running around the playground. The ball kept getting stuck, and one of them would rush back into the class every few minutes to grab a long handled broom from the corner to free it. After about twenty minutes they came back in and plopped down all sweaty, flush with excitement.

AP testing was done; kids like June had taken 3+ AP tests besides mine.

Finals were still a few weeks away.

Teenagers look like adults, but at 16 and 17, they’re still just kids.

Competition heats up fast — rival app Pull My Finger demands fifty thousand dollars for fart infringement.

Within a year the app store is consumed by farts — over 500 of them.

Every tech era rides atop two things: a serious thing — like spreadsheets, search, email. And a delight thing. It might look like garbage at first, but that’s where the winners grow.

Spacewar! Invented computer gaming when a bunch of 1960s grad students got busy wasting the cycles on a million-dollar PDP-1 minicomputer. Teens on AOL Instant Messenger in the 90s started posting updates to their away status buddy lists predating Twitter and all the socials. In 2005, the first YouTube video posted was 19 seconds featuring elephants and their long trunks. Jawed Karim, YouTube founder, called it “Hot or Not for video.”

We’re at an inflection point right now. Enterprise AI is gelling — agents, workflows, orchestration. Business adoption of AI is going so fast the hyper-scalers don’t have enough hyper or scale to cover demand. Everyone is scrambling for capacity. Noise about frontier models is dying down — nobody asks which model anymore. Back in the early Internet did you care what version of the TCP stack you were running?

The AI cognition unit is clear, but where’s the delight unit?

AI slop had a moment, now it’s cooked. OpenAI’s TikTok copy app Sora had a spike of viral growth and six months later they shut it down. Sora hit #1 on the App Store five days after launch, but ultimately burned through $1M/day in compute against $2M lifetime revenue.

That AI slop filling up Meta’s feeds — nobody wants that, it’s garbage. When anyone can create content the value drops to zero. The hunt is on for real vibes, true taste. The junk phase was just an appetizer.

Sure you can vibe code an app in an afternoon and if it solves a problem for you (or delights you) that’s a-ok but good luck stuffing it into the old containers like the app store. They’re not just full, they are fully enshittified — Doctorow’s word — fifteen years of stuffing the stores until discovery died and now everyone downloads the same twenty apps.

Our creative capacity persists, and AI nurtures it — but there's no channel to carry it. History holds another lesson here — creativity will find an outlet, but where? When printing got cheap, we made pamphlets. When video was in everyone’s pocket we made YouTube.

Desire doesn’t expire — never forget the ever popular successor to iFart:

Poop Map launched five years later with a single button UX: Drop the poop, going viral on TikTok in 2021 hitting #1 thirteen years after iFart. Of course iFart is still on the store, still getting press in 2026. Junk has staying power over spreadsheets: delight compounds, while utility churns.

So although the feeds won’t go quietly, they aren’t the place for the future AI fart shocks. Zuckerberg can buy every startup he can find, but bolting stuff onto the feed isn’t working anymore.

AI’s fart app is probably out there already, in some vague vaporous form. Kids messing around sharing creations in private chats. Researchers tiring of training, searching for distraction. Another thing history shows is that it will be something without precedent. Something unintended that we can’t yet imagine.

Farts are stranger than fiction.

Look for something supremely stupid — truly a waste of the most powerful technology ever built. Technologists building serious agent workflows won’t see it coming. Meta will try to buy it, but they’ll miss.

Keep your eyes peeled come Christmas Day 2027.

The big city has been good to me. Coming from LA, I figured I could take on any city.

Nope. But I’ve been trying for forty years, and I started with the biggest one.

###

Swept up in a rolling sea of humanity, I pushed my way across the current, reached the safety of a wall and slid down to sit on my green duffel bag under a big sign, hoping my friend would find me. I couldn’t read the sign, I couldn’t speak the language and the letters made no sense to me.

The smart phone hadn’t been invented.

I was in Shinjuku, the biggest train station in the world. Millions of passengers streamed around me, through a labyrinth of platforms, in and out of hundreds of exits.

I fought to stay awake. The flight from LA took 11 hours. The train from Narita into Tokyo took me another two. The time difference was 16 hours. Tokyo was three times the size of my home town of LA — 30 million people, then the largest metro in the world.1

I was a stranger in a strange land and I went a little crazy after two weeks, hitting the streets on a Japanese whisky fueled vision quest for Oreos, Coke and some sense of normal. But it was the future. We rode the Shinkansen to Hiroshima eating bento boxes screaming past Mt Fuji at 150 mph.

I was hooked. From then on I signed up for every international adventure I could find, and once my kids got old enough we took them too.

I went to Europe to meet JFran’s family for the first time. We had tea in London with scones and clotted cream and strawberry jam. Proper. Rode a red double decker bus cozy in the very back with JFran, ignoring all the sights. Saw Miss Saigon.

Europe became a regular stop once business travel ramped up. Paris, Prague, Barcelona. Grandma roots trips with the kids — Vienna, Budapest, Auschwitz.

European cities are comfy.

Thirty years after Tokyo I got back to Asia by convincing my boss I needed to teach a class in India. JFran joined me for week two, and we did the Golden Triangle: Delhi, Agra, Jaipur. A Sikh named Grover drove us around in a Land Rover with TOURIST stenciled across the front. We were the foreign species and they fed us lots of Biryani.

Delhi topped Tokyo’s strangeness. I can still smell both those cities. Humanity packed in everywhere. India is super dense — four times the population of the US in a third of the space.

To celebrate making it to 50, we went to South America. We started in Cusco, Peru, but this wasn’t about the city, it was about the trail. The Inca Trail to Machu Picchu. My college crew struggled, but fueled by coca leaves and pushed by porters, we made it. Cuy 2 and pisco sours. Time on the trail beat the destination. When we finally got to the ruins it was overrun with tourists and school buses.

A few years later I was back in South America, this time for business. São Paulo is big at two LAs. Twenty million people sprawling across hills that roll through a broad valley surrounded by Favelas,3 punctuated by clumps of high rise, high density apartment buildings jutting from the ground randomly like so many bunches of celery.

And then, I was done.

My interest in new cities waned. Now, the bigger the city, the smaller the world feels to me.

On our honeymoon, we went to Oaxaca. We stayed on the square or zocalo and ate at a different restaurant with different color tablecloths each morning. The second week we went to the beach in Puerto Escondido. We sat on the beach eating cacahuates and drinking cerveza.

Cities munge together for me now, while the other 99% of the world — deserts, beaches and mountains — each feel like its own world.

This week we drove the coast up into Oregon. Oregon, with its moody fog, lush rainforests and broad beaches is different than Northern California’s towering redwoods and craggy cliffs.

When I was 17 and a senior in high school, I took a road trip with three friends. We piled in Dennis’ mom’s crusty tan Honda Civic and drove up the coast to Berkeley to check it out. Once we cleared the LA basin, past the long string of towns to the north, we pulled over at a beach and swung big strands of seaweed over our heads like cowboys.

And guess what? Sam and Dario hate each other.

Hate’s a strong word, and I use it a lot in this piece, but not flippantly. Not a hater. In my personal life I dislike no one, let alone hate anyone. But in my work life there have been more than a few people I hated with a capital H.

But first: Sam and Dario.

Dario came to OpenAI from Google Brain, where he was an AI researcher. A fervent advocate for safe AGI, Dario initially led OpenAI’s safety team before he moved up to run research. Sam’s partnership with Microsoft and subsequent push to commercialize ChatGPT ran against what Dario believed he signed up for. He came to deeply distrust Sam and OpenAI’s direction, leaving in 2021 with a big chunk of his research team to found Anthropic.

OpenAI and Anthropic are both headed for IPOs this year with ridiculous valuations. We’ll be debating the merits of these two companies and their tech for a long time.

While Sam and Dario are the tech rivalry playing out today, they’re far from the first. Take Bill Gates and Steve Jobs. Remember the Mac vs. PC commercials? OpenAI and Anthropic took shots at each other over Super Bowl ads with Anthropic needling OpenAI on their announcement to start placing ads.

But this story isn’t about AI, it’s about how the power of hate fuels the biggest rivalries and the biggest companies and how hate also drives progress within companies.

And it’s not just tech. How about Edison vs. Westinghouse in the War of the Currents. Or Ford vs. Sloan for the future of the automobile industry.

Why are corporate rivalries such a great source of inspiration?

I have some theories.

You pick your friends but you don’t necessarily pick who you work with. Work brings out the worst in people, and the higher you climb, the worse they get. The very fabric of capitalism nurtures this behavior. We dismantle monopolies — it’s all about competition.

Pressure turns coal into diamonds? Not at work it doesn’t; pressure turns annoying people into @$$holes.

If your friend had a rough year, would you put them on a performance plan? If they didn’t quite get everything done you had planned together a year earlier would you put a “does not meet” sticker on their forehead? These people are not your friends.

Once you realize that, don’t sit with it, use it.

Hate has gotten me out of some crummy jobs. Once at a boutique consulting company, my boss — the guy who ran it — was a narcissistic, corrupt tool and his toxic culture pervaded the whole place. Learned some stuff and then I got the eff out.

I hate a lackey. I’ve had a few co-workers that had their head so far up the boss’s @$$ they could feed them breakfast. Add on passive aggressive behavior where they don’t miss an opportunity to take you or your team down a notch. Those guys motivated me to flat out be better than them.

Toward the end of my corporate career there was one guy that was all the worst things. He came over to work for me with a chip on his shoulder because he thought he shoulda been the boss. Insipid, shallow, thin-skinned, self-serving. Man that guy got me going.

Eventually we became peers and you can bet I beat his numbers every single year.

What’s Superman without Lex Luthor? Batman without the Joker? Spider-Man without the Green Goblin?

Now I work for myself. I have no rivals. I only take clients I like. It’s chill not having anyone to hate.

But I’d be lying if I didn’t say hate fueled some of my success. Hate isn’t great, but when it shows up, you might as well put it to work.

Respect the nemesis.

It’s different up north. At the coast you bundle up and fight through the drizzle, climb rocks and search for tide pools. Summer is the cloudiest season. If the sun breaks through, time freezes.

Up here, coffee and chowder beat Coke and Nachos.

Sometimes when I was a kid in LA in the summer, my big sister, whose skin got browner by the day, would let me tag along, and I’d roll down the 10 with her friends to Santa Monica. The 10, AKA the Santa Monica Freeway, stretches clear across LA from downtown, through the basin, to the beach. It’s a ten lane wall of cars that whittles down to four as you approach Venice. Just beyond the Lincoln exit, you pass through the McClure Tunnel and the Pacific Ocean majestically appears, filling your vision.

Now you’re on Highway 1, the Pacific Coast Highway.

###

On our last van trip — seems like forever ago — we went to the Desert.

Driving back from Arizona, I got an email that a local High School was looking for a CS teacher. I stared at that email a long while and I ended up taking the gig. It was fun, but it meant hanging up the van keys for a lot of months.

Since school ended, Alice Cooper’s ‘Schools Out’ has been playing in my brain. I haven’t felt it since I had summers and that’s a long time ago. So, in our first trip back in The Amazing Spider-Van, we went west to the beach coast.

River Road is two lanes from the beginning, and it takes you from Santa Rosa clear across Sonoma County to the Ocean. While the drive takes about a hour, just like getting to the beach in LA, rather than ten lanes of cars cutting a swath through the city, it winds through the county. First, through broad open fields of vineyards. Then, it dips down into redwoods to follow the Russian River through the small towns of Rio Nido (Rivers Nest) and Monte Rio (River Mountain).

The Bohemian Grove, summer escape for wealthy power brokers, lurks in the hills above Monte Rio.

Guerneville is the bigger river town, adjacent to the Armstrong Redwoods and once you make it through there, the terrain opens up again, and you catch sight of the River making the trip to the coast with you. Just past Duncans Mills and then — much like the McClure Tunnel — you round a bend to join up with Highway 1 and the very same great Pacific stretches out before you.

Jenner’s Cafe Aquatica sits where the Russian River meets the Pacific.

Jenner is 500 Miles north of Santa Monica on Highway 1. The 1 ends just north of here, turning into the 101 in Leggett, at the edge of Mendocino county. This trip is a warmup for vanlife summer starting with a longer ride to Oregon next week.

###

The last time we camped at this spot it ended badly. On our last night, running low on food, we went into town for dinner and ate big plates of pasta. Mine had spicy sausage. So good and too much. At 2AM a terrifying, unrelenting, scream came from underneath the bed. The CO² Alarm. We skulked home at 3AM, unable to turn it off.

After consulting with the experts, consensus was our digestion was to blame. Don’t eat pasta with spicy sausage and sleep in a can.

Not this time. After a year of trips I know all the ins and outs of The Amazing Spider-Van.

This trip we slept like babes.

We hit the hills:

We hit the beach:

Salmon Beach faces the full might of the Pacific and it is loud and it is windy. The sand is hot, while the air is cold. It overwhelms your every sense.

The seagulls stare it down:

But it’s a lot, and none of us really know where we’re going to be at the end of the summer.

###

If you’re graduating into this dizzying world of AI, it’s especially hard to make heads or tails of it. When I graduated, I drank a bunch of wine up in the Berkeley hills, and then stumbled down, red faced, to don my cap and gown, walk across the stage and shake someone’s hand. I moved back to LA. Back in with my parents. It took me awhile to find my footing, and longer to find my groove.

once replied to my request for a feature on Substack. She politely told me it’s on the backlog … then left Substack to write about AI. Now she’s immersed in it and has a decided perspective on this:

@jasmine’s substack

🌻 the old world is dying

It’s a scary time to be twenty-two. Recently, I’ve been watching videos of college commencement speakers getting pummeled by boos for talking about AI. I’ve seen the charts showing post-graduate employment rates for engineering majors falling by 5 to 15 percent. And it’s not as if young pe…

Read more

a month ago · 604 likes · 85 comments · Jasmine Sun

There’s a space between turning your back on the waves, and letting them overwhelm you. Take this guy — he’s just looking at something else.

The grind had no color.

Campus was a collection of matching gray rectangles facing off across a concrete courtyard. I worked in building Three. Inside a repurposed conference room (the ‘war’ room), my home was the middle desk in a string lining the wall, each with just enough room to squeeze in a putty keyboard and monitor.

This was the west coast backup datacenter for Bank of America; the bottom two floors of the building were packed with machines. Towering exhaust pipes punched through the roof to vent a battery of backup diesel generators. Should calamity take San Francisco offline, the generators would fire up and Concord would swap in and the Bank would keep cashing checks.

My daily route passed under a faded set of banners proclaiming nonsense like Inclusive Meritocracy and Passion for Winning. Inside the war room, my posse of vendors worked on building out possibly the largest corporate e-mail system of its time and my god it was taking forever.

So, perhaps I was a little tired, tech-weary, tangry1 even when SMS (AKA texting) appeared on the scene as the new tech phenomenon. In the midst of designing the massive distributed storage system needed to keep all the Bank’s email, I scoffed at the 160 SMS text character limit — who’s going to use this?

A couple years later, trying to comprehend the $500 phone bill in my hand and the associated 3,000 texts my teenage daughter had sent that month, I got my answer.

A bright spot from that gray period, was my favorite phone ever, the Nokia 8120. No keyboard. Texting Hello on that phone took over 20 taps. So, no texting. The 8120 existed in a moment when miniaturization was peak and a phone didn’t need a keyboard. It was a glorious device with one job — talking on the phone.

It wouldn’t be long before texting surpassed voice and it all started with a simple hack. Phone networks have a signaling channel to go along with the one that carries your voice. The signaling channel sets up the call, manages handoffs between cell towers, etc. It’s idle most of the time.

A few engineers figured why not use the signaling channel to send system messages? The channel capacity was 140 bytes — enough for exactly 160 characters.2

What can you do with 160 character messages?

Friedhelm Hillebrand did the math. He was working on the new European GSM cell phone standard and he knew the constraints. Friedhelm checked old telegrams, postcards, FAXes. He counted the words, he counted the letters, and yep, turns out you can say plenty in 160 characters. And so it was.

While texting took off thanks to teenagers (not the first, not the last cultural revolution spawned by teens), SMS quickly stormed the world of corporate communications and chat became the way work got done. That was my ticket out of building email systems and into the emergent and colorful world of chat based comms.

SMS became Texting, texting begot Twitter. The original little blue bird was even more constrained than SMS — you had to fit a username in too, so now you had only 140 characters to craft your message.

With Twitter you could reach and build an audience — it was social, it became tweets, chat. In came iMessage, WhatsApp, SnapChat, Slack, Discord. With each iteration, constraints like length peeled away while chat capabilities grew — broadcast, GIFs, disappearing messages, group chats.

Chat is now baked into every tech stack on the planet and of course it’s how we communicate with our current disruptor AI.

While SMS and its subsequent chat brood forced brevity, optimizing for meaning and impact, our latest chat monster, AI, has the opposite problem. AI talks too much.

AI needs to STFU.

AI is built around a different set of limits. Last week, 14-year old Shrey Parikh won the National Spelling Bee with the winning word “bromocriptine.” Shrey didn’t know that word but he knew the pieces: bromo (from the chemical bromine), crypt (from Greek kryptos meaning hidden) and finally -ine (a common chemical suffix). bromo + crypt + ine.

Shrey knows how you win a spelling bee. You can’t learn every word, but if you know the underlying components every single word is made up of, when you get a doozy like bromocriptine, with a bit of info about its definition, origin and pronunciation, you can figure out how to spell it.

AI uses a similar trick. It’s called tokens.

Rather than chunking words out into their piece parts, AI chunks everything. It takes the entirety of every email, every chat, every book, post, newspaper, and blog it can get its hands on, and then, after ingesting it all, categorizes word parts that occur frequently, noting the order they occur.

These resulting word chunks are called tokens. Every response AI generates is assembled by ordering these tokens. While Shrey used semantics to break down and spell bromocriptine, AI uses statistics. Depending on the frequency of the letter sequences across its training data, AI might break it down as: brom + oc + ript + ine.

Tokens are the reason AI famously struggled to count the R’s in the word Strawberry. AI doesn’t see Strawberry as a single word or sequence of characters — it’s a collection of tokens like straw + berry. Tokens are also the reason AI is so verbose. While AI models are great at predicting what comes next; they suck at knowing when to shut up.

They don’t know when to stop, so they over-explain, they just … keep … going. There is always a next token. Bolt on an engagement strategy to keep the conversation going, ending every answer with another question, and congrats you just built the most annoying computer in the world.

Know your limitations.

Since the beginning of search engine time — effectively the beginning of the Internet — Google has been #1. While Yahoo took a run at them in the 90s and Microsoft/Bing in the 2000’s, neither was able to eke out more than a few points of market share.

Google makes gobs of money as the undisputed king of search and they’ve continually used that money to try and find new ways to make money. In tech especially, a company that’s a one trick pony does not last. Google has spent a lot buying AI companies and investing in AI research.

Nine years ago, Google published this paper introducing the Transformer architecture that’s behind every single AI product on the market today. Initially Google didn’t know what to do with it. See, it threatened their core search business, which even today accounts for well over half of the money they make.

While you want to find new sources of revenue — thou must not kill the cash cow. While Google fretted about how to integrate AI into search without killing its cow, two new startups broke the AI market open: OpenAI with ChatGPT and Anthropic with Claude.

These new LLMs started taking traffic away from search. I’ll take an answer from AI every time versus the now ugly, ad-ridden search results Google spits out. So a couple of years ago, Google got serious and shipped Gemini not only as a standalone AI product but also integrated into Google Search. They’re still working hard to catch up, so it wasn’t surprising that last week’s Google conference was 99.9% about Gemini.

Every big tech company has an annual conference and it’s an extended PR event to launch new products. Steve Jobs theatrically orchestrated keynotes at Apple’s annual Worldwide Developers Conference would often end with his tagline: “… and one more thing.” In 2007, that was the iPhone reveal.

Microsoft’s annual conference is called Build and it’s next week in San Francisco. Inside of one of these companies, the months leading up to the event are a mad, frenzied dash to make sure your product is ready for this moment. Make it into the show, and that’s the difference between a product or feature that ships versus one that gets shelved.

If you get into the keynote (the kickoff session on day 1), now you’re living large. Google calls their conference I/O.

Half the stuff at these conferences are hype and never ship as shown. If they do ship, they often disappear because the market doesn’t take to them, so I don’t usually pay much attention. However, there were two telling things in this year’s Google IO keynote.

The first was a variant of Steve Jobs “one more thing ….” Google’s CEO is Sundar Pichai and he’s about as exciting as watching your toaster melt cheese. Sundar opened up the show … Gemini, blah, Gemini, blah blah blah, Gemini.

But then, at the very end of the keynote, AI legend Demis Hassabis, CEO of Google DeepMind took the stage.

Demis dropped this line at the end of the keynote:

When we look back at this time, I think we will realize that we were standing in the foothills of the Singularity.

Oh man. With a general public increasingly freaked out by the looming impact of AI on jobs, quality of life, no datacenters in my backyard … your public conference is not the time to hype the approaching moment when the machines get smarter than us.

Read the room, slow your roll.

The second thing was the return of Google Glass:

Return of the Glassholes. 14 years ago at this same conference, then Google CEO Sergey Brin dashed onstage in cyborg lenses to orchestrate a dance of socially uncomfortable super freaks.

Fail. Nobody bought those things. Creepy. Not to mention you looked stupid and everybody hated you.

Last week’s demo was notably more pedestrian — literally — the scenario was walking directions and ordering coffee. But this is a niche product at best.

At Meta’s annual conference last year, Mark Zuckerberg attempted a more ambitious demo of their new glasses on the big stage. In all my life I have not seen a live demo fail so comprehensively.

Everyone is searching for the killer AI consumer app / wearable. Chat’s nice but there’s going to be a new thing and nobody has figured it out yet.

It’s not going to be the glasses.

Early to bed, early to rise — dough in the belly, dough on the side.

— Been Fryin’, Poor Richard’s Almanack, 1735

School’s winding down and the days are getting hotter. Fifth period hits after lunch and the heat of the day. I unlock the classroom door, muscle my way in before the kids fighting for the roller chairs, hit the lights and crack the back door in an attempt to coax the heavy air to move through a portable with windows that don’t open.

The AP test was last Thursday so now we’re on easy street. Friday I switched out jeans and new t-shirts for shorts and old t-shirts. With the weight of the test lifted, I’m more relaxed and so are the kids. They feel summer coming.

My job is done — they coded their projects, they all took the test and they all did great (so they tell me).

Test scores come out July 6, so we’ll see, but I feel good about it too — we covered a ton of material in the 3 months I had with them. While it’s a constant battle to hold their attention, even when I think they’re distracted they’re listening.

I brought in a bunch of donuts on Friday for a sugar fueled test debrief. Riding high on the wave of finishing the test, I got lots of stories like the kid who got stuck in the wrong test section thanks to the schools intermittent WiFi. Two kids forgot their chromebooks and the administrator had to sprint to the office and back to get spares for them. I gave her a donut for her troubles. A lot of kids finished early and just had to wait (the test was 3hrs).

One kid kept sneaking back to the boxes as we talked, eventually downing four donuts. He was a little green by the end of class, but then went on to catch for the varsity baseball team. They won league later that day confirming donuts are a critical element of any diet.

But I can’t let these kids go without hands-on AI — it’d be a crime to teach CS without AI today, so this week we get AI to write the code.

Last year, vibe coding became a movement powered by tech enthusiasts thrilled with AI writing programs they wanted but couldn’t code. My kids had never heard the term, and I don’t like it much, so I’m calling this unit AI Mad Libs. They all know Mad Libs and that surprised me — they aren’t always on their screens. I build short AI prompts for various games, they fill in the blanks to make it theirs and we iterate from there.

Their ideas are better than mine. Honestly they do better when they go rogue off the prescriptive prompts. AI is good enough to make it work. One of my quiet kids yesterday sat in the back of the room, heads down building Neon Parkour. He had a crowd around him playing it by the end of class.

We’ve also got: Grand Slam Guesser, Forge of Legends, Echoes of Time, Aura Chiba and Neon Sphere 3D. Still got one more day.

Check them all: https://analynet.xyz.

I had an ulterior motive in bringing in the donuts — I wanted to show off my demo AI app — DOUGH DROP.

With Dough Drop you build your dream donut, watch it bake and then get roasted by a gallery of top AI Celebrity Chefs: Jacques the French pastry chef, Guy Fieri (Sonoma County’s hometown hero) and Gordon Ramsay.

You can share the roasts with your friends. Guy hypes you up while Gordon tears you down (usually): 1

In the midst of the Friday’s donut-palooza I checked in with the senior table. We’ve been through it and they’ve decided I’m ok now. They asked me if I knew anything about stocks.

I try to figure them out; they try to figure me out. I said sure, I know a bit. Then they asked me about day trading. I said, yeah if you like gambling. They like gambling.

One of the seniors was serious about this, so we had AI build this Python WALL STREET SIMULATOR and we ran through it as a class.

It doesn’t matter if you have a lot or a little to start. Put in the minimum to open a brokerage account, throw a little bit in a month. $10/mo is a start. Get a job, up it to $100/mo. Get a promotion, up it to $1000/mo.

This is the thing I will never convince these kids or any kid of — they have the best investment advantage right now — TIME. If you’re under 30 reading this — do it now! At 16ish my kids have 30+ years and a little bit of money will grow into a big bit of money with some time.

As far as what you invest in, as long as it’s a diversified fund, that’s secondary to just getting started. Don’t get stuck on picking the perfect fund. Sure fire winner — drop it in a no-load broad market index like VTSAX and just keep watering it.

Market conditions? No matter. You can’t control the future. With a time horizon of 10+ years it’ll even out.

Graton Casino down the road here just added a new wing for high rollers including fancy restaurants like SoCo Dough Co. They’re selling this $275 Donut:

Nah.

What do they even do in the bathroom?

I’ll never know, but I do know one thing — not a single kid in my class uses Facebook.

I asked them earlier in the week teaching them how all this AI stuff even happened. They do use Instagram, an app that was perfect in its time. The smartphone had just blown up, so everyone was walking around with a marginal camera in their pocket. Kevin Systrom was the guy who invented Instagram. His first App was called Burbn1 and it was for check-ins.

Burbn wasn’t doing so great but people liked posting pictures so Kevin went minimal — one app, all images. App written, he went on a break to Mexico with his girlfriend. Nicole didn’t think their phone vacation snaps looked good enough to share. So Kevin went back to the hotel and built the first filter. He called it X-Pro II and it’s still in Instagram today.

The next day Kevin posted the first Instagram photo from a taco stand (that’s Nicole’s sandal next to the dog):

A star was born. Kevin caught lightning in a bottle. Instagram grew to 30 million users in 18 months.

Facebook — Mark Zuckerberg — saw the threat. Facebook’s mobile app was worse than the crappiest of crappy phone cameras. Instagram’s app was simple, elegant, mobile only and it was beating the pants off FB: 2

So Mark bought it.

He courted Kevin with a combination of money ($1B) and threats. He promised Kevin Instagram would stay independent and he would still be in charge.

Of course that didn’t last — Instagram was growing faster than FB — so Mark took it over, crammed it with ads and a few years later Kevin left. Don’t feel bad for Kevin — choices — he left a rich man, but he lost his creation.

Same story, different ending: Snapchat. Evan Spiegel invented Snap, and when he was 22 Mark came for him too — offering him a mind boggling $3B dollars for his company. But Evan said no. That was the day I started using Snapchat. I remember reading that headline and thinking Evan was my man.

Mark holds a grudge. He couldn’t have Snapchat so he blatantly stole every innovation Evan came up with and he came up with a LOT. Snap was a revolutionary app in its prime: ephemeral chat, stories - yes Snap invented stories! Mark also stole Spectacles — now creepy Meta AI glasses — he’s still trying to get that one to work.

Through it all Evan held on and while the future of Snap is not clear to me — one thing that is clear is that Evan is still running the show.

Mark hasn’t come up with a single original idea since he created Facemash in his Harvard dorm — a hot-or-not website to rank college girls. He’s an ultra competitive, cutthroat businessman and boy did he nail ads, but he’s got no heart, no innovation, no soul.

Mark created an ‘insider’ class of Facebook stock and he owns most of. It has 10X the voting rights of a standard share, so Mark controls the board and can never be fired. With the ad money making machine and no accountability, Mark has gone off on some ridiculously expensive attempts to invent the next thing.

Take the Metaverse - he renamed the company Meta and spent over $80B building it only to shut it down earlier this year without making a dime.

Even though my HS kids are on Instagram, they aren’t sharing pics. IG is no longer a place to share with friends. That happens in private chats now, across messaging apps like iMessage, Signal and WhatsApp.

For instance, here’s Boo. I am fortunate to be treated to steady diet of Boo pics in our family chat. While our dog Lola had her own Instagram account, that was a million years ago.

Boo knows better:

Mark see’s this happening. Another big threat. He may be a tool, but he’s not a fool.

So, Mark bought it.

Shockingly once again he promises WhatsApp’s founders — Jan Koum and Brian Acton — to preserve the core elements of the app they built (privacy, security, ad-free) and that they will be in charge. The real shock is after what happened to Kevin and Instagram, Jan and Brian fall for it. Guess it’s hard to turn down the big money. Well in came the ads and they both left abruptly a few years later.

AI is bigger than the mobile wave. AI is bigger than the messaging wave. AI is the biggest wave and Mark see’s it bearing down. What’s he doing about it besides shoving AI generated slop into every feed possible?

How about laying off a bunch of people and then telling the ones left they have to run AI training tools on their computers that record every keystroke and also screenshot while they’re working? When jobs went overseas in the 2000’s there were lots of stories of training your replacement on the other side of the world.

How about training the clankers to take your job?

Meta’s AI model Llama was hyped as open source but the industry saw through that and then and then their latest model Llama 4 well — it sucked worse than those first mobile cameras.

Since not much of this is working, Mark is back to his old playbook of bullying his way into a new market by buying every AI company he can get his hand on.

That’s not going great either.

There’s a big tech trial going on down in Oakland this week. Elon Musk is suing OpenAI for the way they converted the company from a non-profit. Satya Nadella Microsoft’s CEO and the last boss I had testified on Monday.

A bunch of juicy docs have come out of this trial including these texts between OpenAI’s Mira Murati (then CTO) and Sam Altman (CEO):

Sam did indeed get fired 3 and Mira was thrust into the spotlight as OpenAIs CEO for 5 days.

Mira is legit. She’s an immigrant from Alabania and so smart. If you ever wanted to understand what’s really powering today’s AI, Mira explains it:

After the Sam ordeal, Mira left OpenAI, and an elite group of AI researchers from OpenAI went with her to form a new AI company called Thinking Machines Lab.

Here’s what she said when she left:

So Mark bought it.

Not this time. He sure tried — he came after Mira and her new company, but Mira said No. Mark went nuclear, headhunting every researcher in her company.

They ALL said no.

Last week Meta announced a dip in their users numbers for the first time in, well in forever. They’ve never lost users. I think it’s a sign and I’m not the only one:

Meta is Dying. It’s About Time.

Facebook will not go quietly. Not only because Mark but because network effects. Where else can I go and hang out with an unholy hodgepodge of friends from elementary school to my high school girlfriend to random people I used to work with? I still have to open Facebook once in awhile, and I hate it.

The future I’m looking forward too, is one where we’re done with Facebook, and Mira and her ilk build new ways for tech to serve the humans.

My parents lived through D-Day, the Cold War and NATO, the civil rights movement, Watergate, the dissolution of the Soviet Union. Technology blossomed around them from the transistor to the Internet, to a computer beating a world master in Chess.

TALKIE — a new ‘vintage’ AI large language model doesn’t know any of this happened.

TALKIE was trained on over 2 million books and periodicals and journals — every single one published before 1931.

Why 1931? Because any publication over 95 years old is public domain — there are no copyright restrictions.

How is such a huge corpus of paper books available to train a machine? While the technology to convert a printed page into computer bits 1 has been around since the 1950’s, Google supercharged it in 2004 in an effort to digitize every book ever written.2

Google created custom hardware powered by the latest tech that sped through books, scanning 1,000 pages per hour, but error rates — especially on handwritten text were still high. In 2009, an acquisition solved that problem.

The acquisition was reCAPTCHA. That’s those annoying puzzles you get on websites to prove you’re a human. reCAPTCHA put the words Google’s OCR failed on in front of millions of humans to solve. Those crowdsourced answers were built into the scanning algorithms which gave Google the final piece it needed to unlock digitizing old books at scale.

TALKIE is rooted in the tech of its time, and in 1930 that was Radio:

Why build TALKIE? To see if AI can predict the future.

Demis Hassabis, CEO of Google DeepMind came up with the Einstein test earlier this year as a way to gauge if AI can ever actually be “Intelligent”:

TALKIEs creators want to know the future. Specifically, they want to know if a machine can predict the future.

TALKIE doesn’t know about what’s happened since 1930. But we do. Can TALKIE predict what we know happened?

While radio was the transformative tech in the 1930s, movies were becoming more popular — especially ‘Talkies’ — the silent movie killers. But what about the visual equivalent of Radio — Television?

I asked TALKIE if we’d ever see moving pictures through the air like Radio:

That’s a natural scientific arc. But what about Humanity? I asked Talkie if Germany could become a fascist state:

Eleven years after TALKIES 1930 cutoff date, my dad was fourteen and could watch TV dispatches from Europe about WWII. TALKIE was right about the technology.

Eight years after TALKIES 1930 cutoff date, my mom was twelve and fleeing Austria on the Kindertransport. TALKIE was wrong about Germany.

Predicting or creating novel science — I can see that. Predicting the foibles of humanity — I don’t think so.

They sure did a lot of damage though.

I got stuck in the middle of it and it was utter and complete chaos — the onslaught of worms, viruses and malware was nonstop, and the tools we needed to fight them didn’t exist. Melissa, SQL Slammer, Code Red, Nimda, Blaster and of course … ILOVEYOU.

Then there were the guys who saw the wave — saw the opportunity — fashioned their white hats, grabbed a surfboard and rode that wave. While they promised to save us all from the scourge — they made out like bandits.

Still, it ended badly for many of them.

My gig back then was building massive corporate email systems. Custom systems built up around a product called Microsoft Exchange. Exchange 5.5 had built up a loyal corporate following and the new release — Exchange 2000 — would go on to become the cornerstone of Microsoft’s enterprise collaboration business.

PCs had blown up and they were on every desk, in every classroom, every warehouse. The mantra was make it faster, make it easier, make the components and the software easier to use.

Security was an afterthought, an annoyance, something that slowed down the pace of releases.

ILOVEYOU took advantage. A few lines of Visual Basic written by a frustrated college student in the Philippines, took down the Pentagon, the FBI, the UK Government, and AT&T.

Here’s the guy that wrote it - Onel de Guzman.

Onel’s college professor threw him and his thesis project to steal passwords out of school. Unethical he said. Frustrated, Onel went home and wrote up the code. He couldn’t afford internet access outside of school and this one trick was going to get him a stash of dial-up credentials to keep going.

He attached his creation to an email, typed a line to a friend in Singapore, hit send, and went out to get dinner.

That one email spawned a global crisis. In 10 days ILOVEYOU took over computers around the world, leaving behind a trail of destruction. 45 million infecting computers, hundreds of millions of deleted files, tens of millions of stolen passwords. ILOVEYOU scraped so many credentials, the email accounts de Guzman sent them too became hopelessly backed up and inaccessible within hours of that first email.

Antivirus software was nascent at the time — there were no techniques to counter a zero-day attack like this.

Clicking on the love letter (which appeared to be a simple text file, thanks Windows)1 not only grabbed cached passwords, it replaced every picture with a copy of itself and then mailed itself to everyone in your address book. Who can resist reading a love letter from someone you know?

Exchange couldn’t keep up with the mail volume. The only solution was to take down every server, purge the mail queues and manually patch every infected computer before bringing the servers back up. On a large system this took days. I worked some long nights.

Onel later said:

Peter Norton and John McAfee had established their tech business’ before ILOVEYOU. Norton came from a comfortable family in Seattle, and after getting laid off by the aerospace industry, he became a buddhist monk in San Francisco.

Then, he accidentally deleted a file on his computer. Rather than anger, Peter choose the path of the buddha and instead of recreating the whole file, he wrote a program that recovered it. Then he wrote a few more system programs. He started dropping into computer user groups and Norton Utilities was born. Norton Antivirus came soon after. Business was good; ILOVEYOU made it exponentially better.

There was a time when you couldn’t walk into a computer store without seeing shelf after shelf of yellow boxes with this picture of Peter on it. Norton AV was what you bought to protect your grandma’s PC from the Onel’s of the world.

John McAfee’s origin story is rougher. Born on a US Army base in England to an American father and British mother, his dad abused him and his mom, eventually committing suicide.

John got his degree in mathematics and bounced around as a programmer in the defense industry including time at NASA. When someone showed him one of the first viruses on a floppy disk — he saw opportunity. It’s all just software.

John wasn’t the safe choice like Peter - McAfee weaponized viruses; he sold fear. McAfee AV was the only thing that could save you from the scourge of software coming for you.

Like Norton, ILOVEYOU was a gift to John McAfee. McAfee and Norton Antivirus traded places as the #1 and #2 AV companies, minting money in the process.

In January 2002, Bill Gates realized the scope of the problem and slammed on the brakes, pulling 13,000 Windows engineers off feature work to focus 100% on security. Every developer in the company underwent rigorous security training. I spent a week up in Redmond deep in it myself.

Peter, John and Onel’s malware wave ran out of steam as each new version of Windows shipped with a lot less bugs and a lot more security baked in. With AV built into Windows, the market for standalone AV software evaporated — bad news for Peter and John’s companies. 2

But Peter and John were smart — they both cashed out early. Peter left at the top of his game with 70 million, becoming a philanthropist, art collector, and building his dream home on cape cod. While his face kept selling product, he never looked back.

He’s still kicking.

McAfee left a few years after Peter with $100 million — he would later call McAfee’s software ‘the worst on the planet.’

But John did not go quietly and collect art — he moved to Belize, living a life surrounded by big dogs, drugs, guns, and murder. His behavior became increasingly erratic and paranoid. His neighbor poisoned his dogs; his neighbor was found executed. Fleeing Belize from the murder charge, he returned to the US, embracing crypto while mocking the IRS saying he would never pay taxes.

There’s a ton of media on John if you’re curious; this video he made is a glimpse of the madman.

He’s dead.

The IRS caught up with John and he fled the US, living off his mega yacht with his third wife, four guard dogs, seven staff and two security guards. He was made land in Europe and was arrested in Spain, charged with tax evasion, money laundering and fraud. He died of suicide in 2021 at age 75. The day the Spanish Court ordered his extradition back to the US to face charges.

What about Onel? It wasn’t hard to trace ILOVEYOU back to him. His email and hacker handle headlined the code. But the Phillipine’s had not written any cybercrime laws yet and he walked. However, the attention was too much for him and he disappeared.

A reporter caught up with him a few years ago — now he fixes phone screens in a kiosk in a Manilla mall.

Who would you choose - the Monk, the Misfit or the Madman? No one wants to be Onel. Opportunity wasted. I choose Peter. Leave at the top; live your best life. When I told my HS kids this story, a lot of them wanted to be like … John.

Here we are in a new wave. Who are the Peter’s and John’s and Onel’s in today’s AI wave? There’s plenty of characters to choose from. Sam Altman, head of OpenAI and Dario Amodei, head of Anthropic are easy picks. Neither is a Monk; both may be Madmen.

The closest we’ve got to Onel (so far) is Australian Vibe coder Peter Steinberger — creator of OpenClaw. I wrote about the recent riot Peter caused across the globe in Agent Smith. But unlike Onel, Peter played a smart game, landing a cush gig at OpenAI along with a fat paycheck.

But we’ve only just begun. Millions are out there hacking away at AI as you read this — everyone looking for their moment. Keep your eyes peeled — we won’t see this wave for what it is until we’re looking back.

But I know bridges. I look for them.

We drove over it in the morning and it was a beauty:

Growing up in LA, I didn’t see many bridges. So for my high school science project, I hunted one down, driving an hour south to San Pedro for a date with the chief engineer of the Vincent Thomas Bridge.

He took me up on the catwalk below the bridge. Glorious and dizzying and noisy. He told me the bridge is so big the painting never stops. When they get to the far side they just start over again.

Maybe bridges were part of my motivation to head north from LA up to the San Francisco Bay Area where I could spot this beauty every day. 1

The Bay Area is defined by its bridges — the Golden Gate crosses the mouth of San Francisco Bay and there are four that cross the Bay. From the top of the bay to north to south: the Richmond, Bay, San Mateo and Dumbarton bridges.

In Sneakers, Robert Redford and his crew sleuth out the bad guys by doing blind bridge math. Pick the wrong bridge during rush hour, there goes your evening.

I had graduated and moved back to LA when disaster struck. The Loma Prieta earthquake not only pancaked most of the 880 through Oakland but a massive section of the top lanes of the Bay Bridge (where you drive heading into San Francisco) fell onto the bottom lanes which head back to Oakland.2 I watched it go down on TV along with millions of others just as the 1st inning of the Giants versus A’s “Bay Bridge Series” was getting underway.

CalTrans patched up the old span, building a new bridge side-by-side and then finally blowing up the old pilings. I was commuting from Oakland into the City for a good part of that and watched it happen. Two decades and $7B dollars later the new eastern span opened, crowned by a massive 500 foot tall tower made in China.

A few years ago I biked over the new span. We met in Larkspur in the Ferry Building parking lot and first went over the Richmond bridge. That was not fun. You’re biking into traffic in a makeshift bike lane with temporary barriers —honestly, it’s terrifying.

But the new Bay Bridge has an official bike lane that’s big enough for double wide strollers and dogs.

It’s a climb to get to the top of the span. From the top we could see the ferry pulling into Treasure Island. You can’t bike across the old span so we needed to catch that ferry. We didn’t — it pulled away just as we made our way off the bridge.

The view wasn’t bad waiting for the next one.

The Ferry is bike friendly.

From the Ferry Building we biked down the Embarcadero, through the Marina, and joined up with a throng of tourists in Golden Gate Park biking and walking over the Golden Gate bridge.

65 miles and six hours later — we got back to Larkspur where we started.

Three bridges and a ferry.

I went back to LA after graduating from Berkeley, but I only stayed a few years before I moved to another water city — Seattle, the Emerald city — home to more beautiful bridges.

To get from Seattle to Redmond, you have to cross Lake Washington. It’s too big for an arch bridge or a suspension bridge. Enter a third type of bridge — floating. Once I drove over the old floating bridge in a storm. It was bucking like a sea serpent; I figured that was the end.

They’ve since rebuilt it and it’s the longest floating bridge in the world.

Bridges aren’t easy. For a minute, the Tacoma Narrows Bridge was one of the longest suspension bridge in the world. Until a bit of wind hit the perfect resonant frequency that amplified the structure with every gust and blew the bridge down: 3

There’s one more type of bridge — and that’s a cable bridge. My hotel in São Paulo looked down on a fine specimen of this on my last International business trip.

This is the Ponte Estaiada bridge in São Paulo. Rumor has it I returned from Brazil in 2020 and brought COVID with me.

After I left Microsoft, we moved up to the Sierra Nevada mountains, just down the street from the famed Donner Summit Rainbow Bridge on old 40.

Many a car commercial has been filmed here. That’s Donner lake below. The rainbow bridge is another variant of an arch bridge. I’ve also biked this bridge down to Donner lake and back, but it’s sketchy down and torture up.

Sonoma county north of San Francisco is criss-crossed by rivers heading to the Pacific. A lot of the bridges — like Wohler bridge here — cross the Russian river.

We rode over Wohler last month, just before they closed it for repairs. My bike was feeling the miles and needed some work too, but I got it back from the shop yesterday so now it’s time to find more bridges.

AKA “the whole story.”

Mythos is the name of Anthropic’s latest AI model. Unlike their prior models released with fanfare and hyperbole, they’re not releasing Mythos (at least not yet).

They claim it’s too dangerous.1

From Anthropic’s national security team:

Why oh why would a company build a thing that could set back cybersecurity 30 years, take down the internet, and force a rewrite of every piece of software?

Put a pin in that for a minute and lets talk about OpenBSD.

The B in OpenBSD stands for Berkeley. It’s one of the early Unix 2 Operating Systems, and the very same one I learned to program on when I was a wee sophomore at Cal.

I spent a lot of time down in the basement of Evans hall.

The computer lab had no windows, lit instead by a matrix of florescent tubes, their harsh white light bouncing off row after row of black and white CRT terminal glowing screens. The floor was linoleum and slightly sticky. The lab was powered by all-nighters fueled with caffeine to a Motown soundtrack courtesy of a crowdsourced boombox my professor bought and chained to one of the desks. 3

The room smelled of sweat and — as the semester progressed — a tinge of desperation.

Our programming stack was housed next next door in a room full of mini-computers running OpenBSD Unix.4 We edited our code in this super sexy editor: 5

After a few hours down in that basement you could feel the hummmm of computing within you, around you, through you. You were in the belly of the beast.

Today, in my high school class, the stack is decidedly more abstract. The kids write their programs in an online educational program called Codio. Their code runs in a virtual sandbox on a nameless server up in the cloud in a datacenter far-far-away. The classroom has windows too.

The fundamentals of programming are not different than those I learned forty years ago. But I have to work hard so the programs they write seem even a little bit relevant to what they see on their phones, their laptops, even the school’s website.

I think they have similar real world disconnects across most of their classes. It must be tough — they certainly aren’t of an age where they are just seeking knowledge.

It’s no wonder they’re easily distracted.

A lot of my kids pick up a Rubik’s cube the minute their hands come off the keyboard. I’ve learned they’re still listening to me. Better than they would without the cube. If you look into the why of this, you’ll find Neurodivergent.

I probably fall under this category too, but who knows? Brains are weird.

Also amazing, so we started building machines to recognize patterns and create things mimicking how we think. The first AI approaches were grounded in traditional programming. Now called symbolic AI, they use logic and rules along with purpose-built knowledge structures to make computers reason step-by-step.

Today’s AI is markedly different and it doesn’t always work.

If you’ve talked to an AI chatbot you know the responses can be wildly variable. It’s part of the novelty and the fun. It’s a prediction machine built on statistics. Two runs with the same exact input can have very different outputs. Hallucinations and inaccuracies prevail.

JFran uses ChatGPT for lesson plans. It’s good at cranking out say “Ten fill-in-the-blank adjective sentences for a third grade bilingual class in a nature unit”

But then she tried to get it to do a simple Spring Crossword puzzle:

If you ignore the paint-it-black on the right, it does resemble a crossword puzzle. But then you realize the words don’t fit and the numbers don’t match and it’s just all a terrible jumble that makes my head hurt.

So I tried. This time with a paid subscription and the latest thinking model. I asked for a simple 5 x 5 using computer networking terms. After 14 minutes and 17 seconds of “thinking,” I got this:

Before I started I asked ChatGPT if it could pull off a crossword puzzle:

I questioned it’s confidence:

It got shy:

This is the madness of Generative AI Large Language Models.

The AI labs introduced thinking to manage this behavior. That’s where the model interjects rules and stepwise processing to evaluate results before proceeding. As you can see above that doesn’t always work.

Now we have skills and projects where you can memorialize the steps after a successful session/task completion for repetition and keep a running record of resources. This works well for defined tasks without a lot of variance. I use a version of this in my class with a model that’s constrained to operate on a set of curated sources. In my case all the official AP curriculum.

While this is getting better, it’s still not great. There’s a debate between two opposing AI camps: neural networks (statistical predictive models) versus traditional machine learning (symbolic deterministic models) for the future of AI.

Hybrid models are now showing up and it might be the best of both worlds. Think of how you or I might balance our intuition with evaluation when making a decision.

Recently Anthropic, the same company that is scaring the pants off us with Mythos, accidentally revealed a bunch of their source code on the internet (hehe). It was Claude Code (code) — a very popular AI coding tool.

Analysis of the leak pointed to a library with a large and complex decision trees based on traditional IF-THEN statements. So it appears Claude Code is a hybrid architecture combining new and old AI techniques.

While I only have passing understanding of all this stuff, one thing I know is the best architecture eventually wins. The example of this I’m teaching my kids right now is the Internet. Out of tons of networks the set of protocols that define the Internet won. That took about twenty years.

Maybe we regroup from developments like Mythos and realize humanity doesn’t need the risk of powerful computing systems that are a black box to us. We need a deterministic layer to make sure certain rules are always followed.

Like don’t break the Internet.

You build a program, let’s call it Halt. It isn’t complicated, Halt simply determines whether another program will run forever in an infinite loop or it will eventually finish.

But then someone goes and writes a program called Trouble that looks at what Halt comes back with and does the opposite. So if Halt says Trouble will loop forever, Trouble does the opposite thing and finishes its run. If Halt says Trouble will finish, again it does the opposite and loops forever, never finishing.

Trouble is a perfectly legitimate program. It may not be trivial to write, but programming is all conditions and loops.

So what happens if Halt tries to evaluate Trouble? It can’t work.

Halt cannot exist. The Halting Problem cannot be solved. In CS lingo it’s undecidable.

Alan Turing came up with this around 100 years ago just as we were building the first computers. This is from the same guy who created the Turing Test, one of the more interesting questions in computer science until AI and Large Language Models came along and showed us just how easily we can be fooled.

As opposed to the impossible mind-bendy halting problem, impractical problems are ones we’d very much like to solve. But, the algorithms to solve them require so much compute that before we get an answer the sun will become a red giant and turn the Earth into a tater tot.

The classic example of an impractical problem is The Traveling Salesman. Again the problem is deceptively simple:

A salesman needs to visit 20 cities to hype his new AI app (which looks like juice boxes). What’s the most efficient route?

We solve this problem by checking each possible route. If the salesman is going to 5 cities you can do it. There are 120 possible routes. The math that counts all routes is 5 factorial or (5!).

5 x 4 x 3 x 2 x 1 = 120

You’ve heard of exponential. That’s growth by a constant factor, for example doubling every year. Perhaps you’ve also heard that when AI starts building AI it will grow exponentially smarter until … the SINGULARITY.

Factorial growth is significantly faster than exponential growth — every new city you visit increases the number of routes by a higher factor. Say the AI juice boxes are taking off, so now our salesman has to visit 10 cities. At 10 cities, the number of possible routes jumps from 120 to nearly 4 million.

Twenty cities? That’s over 2 quintillion or 2,432,902,008,176,640,000 possible routes.

The Traveling Salesman Problem isn’t impossible like the Halting Problem, but it’s impractical. Forty cities? It would take the world’s fastest computer hundreds of years to calculate every route. That’s really not so many cities — no juice box for you! In computer science, that’s called an intractable problem.

A heuristic is a clever workaround or hack that gets close to the right answer. For the TSP, just pick the closest city each step of the route. You won’t get the optimal answer but it’s close enough. That’s how your Amazon driver figures out where to drop off your monthly supply of Hanes boxer briefs.1

I wrote a program that does the math — you can play with it here. 2

At the turn of the century, math enthusiast and businessman Landon Clay launched The Millennium Prize Problems. Seven unsolved problems. Solve one, win a million dollars. The TSP lives in one of these.

The world is full of problems to solve that would do more than deliver juice boxes. Or boxer briefs. Making progress on these could be big — we could live longer, happier lives on a healthier planet.

So how’s our great new innovation AI doing at solving these hard problems? AI isn’t going to save humanity by doing your homework or drawing weird pictures. Our AI salesman will tell you it will improve everything you do just a little bit. You’ll become more productive.

Every product I ever worked on promised this same thing. It’s impossible to quantify.

AI has taken a couple of swings:

Google’s AlphaFold made a major breakthrough in protein folding, accelerating biological research (like understanding Alzheimer’s) leading to a Nobel prize.

Another: AI helped decipher the carbonized Herculaneum scrolls that burnt up when Mt. Vesuvius erupted 2,000 years ago. So far they’ve found the word ‘purple’ and a discussion about the nature of ‘pleasure.’ Translation is ongoing. I’m certain more p-words will be found.

Perhaps we might also work on, say, the climate, Alzheimer’s, food for everyone.

How about we figure out fusion and cure cancer?

I don’t have a quintillion years.

The world feels a little wobbly, and so does the job market. AI is a big part of that.

Up until now I got by on my spidey-sense. I always got out before the company went down, or the bad boss imploded.

Often, just in the nick of time.

Dysfunction junction? Plan your exit, pick a better boss, get a mentor, find some allies.

Ego-maniacal boss? Toxic workplace? Manage up, channel that anger energy into finding a greener pasture. You can’t get out of this one fast enough.

Company struggling? Work connections, build bridges. Get out before the ship goes down.

One rule covers all of it: keep your head up. There are always clues but you have to be looking for them.

What’s going on with staff, co-workers, management? Are good people leaving? Managers/Leaders leaving or playing an odd game of musical chairs? Is the company in a hiring freeze or not attracting strong talent?

Time to look elsewhere.

Re-orgs are great — full of clues — who got promoted, who got a larger remit, who got a bigger budget? Go work for that group. Network like crazy on the inside.

Same with products — what’s making money for the company? What’s not doing so well or getting de-prioritized? Go work on the hot stuff. Better to be in the hot seat than no seat.

Getting fired is plenty unpleasant. I’ve had brushes with it one way or another and while there are plenty of euphemisms — it’s always personal — even if your company or your boss tries to make it impersonal.

Take RIF for example — Reduction In Force. The Force made me do it.

Here’s another awful acronym: DNM (Did Not Meet). Get one of those on a performance review, turn the networking dial up to 11, start looking outside the company at least a little bit every single day. Start hoarding office supplies.

It isn’t you. It’s one of the above things, but none of that matters if you’re out of a job. To keep your agency — your ability to navigate and manage the ever changing tides without becoming a victim — develop that spidey-sense.

So how did I get fired? By taking an emergency job in a public High School in a district, in a county — in a state — where public education is under duress post COVID along with a myriad of other issues I don’t pretend to understand.

You can read about what’s happening at my high school here. This cut: Information & Communications Technologies - Computer Science .2 FTE 1 — is the position I’m filling through the end of the year.

I’ve been whining about how much work this .2 FTE is, so while I’m having fun (mostly) it’s messing with my semi-retired vanlife, so it’s all good.

Reading the workplace was tricky enough and then AI entered the picture.

I may be off the career hamster wheel, but I’m still survival-coded to try and figure out the next thing. So, while my livelihood no longer depends on it, I’m intensely AI-curious.

I happily jumped into this teaching job even thought I’d never done it, knowing Claude2 had my back. A couple weeks ago I wrote about using all the AI Google offers in the classroom in AI High.

Last week we finished our big coding project so it’s time to pack the next six weeks with all the knowledge they need to kill it come AP test day.

Saturday I spent some quality time with Claude to figure that out.

I fed Claude the school calendar, my M/W/F schedule, teaching styles that land with the kids, what they theoretically covered before I took over 3, what the AP College Board says they have to cover, and what’s actually on the test. We went back and forth until we landed on a plan I like.

I won’t go back to the online course (which the kids hate), so now Claude (along with Gemini/Google Classroom) works with me to build lesson plans grounded in this larger plan on the daily.

If you look at AI as just the latest tech, it’s not so different from picking up on the next hot thing. But AI isn’t just the latest tech. This cognitive layer we’ve built with a natural language interface is quite different from everything it’s built on top of lower in the stack.

A constant through every job I’ve had — including my current HS teacher stint — is I never have enough time to do all the things I want. The more engaging the job, the more I’ve found this to be true. If you can compartmentalize AI to take on the stuff in your job you hate — the grunt work — well that’s one way to get to know it and make it work for you.

One gauge of the impact AI might have on jobs is the first industry it’s completely upended: software development.

Anthropic (makers of Claude) says AI now writes 90% of new code. But they’re still hiring software engineers, including entry level. They look for ‘AI natives’ that have embraced AI from the outset and consequently have a feel for how to work with it, imposing judgement, taste and design skills to build great software not slop.

They’re preaching — AI hasn’t killed the job of a software engineer (yet) but it has fundamentally changed what the job is.

My HS kids will soon be out in the wild dealing with this uncertainty. I feel for them. My own kids are challenged on the daily by their companies to use more and more AI which is another level of overhead when you’re already busy. I don’t envy my buddies still at Microsoft refactoring their jobs around AI while the tech stack constantly changes along with the work flow.

Every tech company in the world lives in this pressure cooker, fighting to stay relevant in a world getting eaten by AI. There will be winners and losers.

But that’s the deal, so better run towards it, spidey-sense ready, rather than run away.

My students were pretty excited for a break too, but now we’re back at the grind.

On Monday, I get to school a few minutes early: open up the classroom, unlock the laptop vault and connect my laptop up to the big screen.

We have to get the coding project for the AP Exam done this week. I have two more units yet to cover plus review and exam prep.

We’re short on time and I‘m herding cats. Teenage cats.

The first bell rings and one of my seniors wanders in with a pizza box. Fifth period is right after lunch. I don’t know what this kid does during lunch but apparently eating is not it. In addition to bringing his lunch, this kid brings a welcome dose of seriousness to the senior slacker table. He’s worried about grades. We talk.

The rest of the kids straggle in; chatting, grabbing laptops from the vault, settling in with their table tribes. My class is in South Campus and it’s a long walk here.

I mentally take roll — everyone’s here but one kid and he’s a superstar (coding the NYT’s Wordle game by hand for his project) so I’m not worried about him falling behind.

I go through slides covering online project submittal. It’s a lot: full project code, a video showing the program working and a set of key code blocks.

The senior table is increasingly distracted by the pizza. Talk of sharing gets louder. It’s pepperoni and it smells good. I’m fine with the pizza but don’t talk while I’m talking so I stop until they close their laptops and their mouths.

A kid wanders in late, and sits with his head down on the desk. He finished his code before the break so I let it go.

I get through most of my spiel before I start losing them. One of my 10th grader’s who looks continually lost starts with the BIG yawns. For directions like this, their attention span is even shorter than normal, so I cut it short and tell them to get to work.

I circle through the class, checking in with each of them. I have three girls and they sit right up front — a 10th, 11th and 12th grader. They shelter together across grades, avoiding the chaos around them. They’re consistently my strongest group and keep me on my toes.

I make my way over to the solo-table loner in the far corner. While he barely makes eye contact with me and hates group work; he’s generally on it.

Next, the junior table. After the senior table, they’re the loudest because they play the games as they build them. One of these kids is on Varsity Baseball so he’s missed a few classes, but he show’s me this marvelous geometry run game he’s building, so all good.

That leaves the table between the girls and the seniors. A mixed group and perhaps my favorite table. This is where the kid who came in late and immediately put his head down is at and now we’re going on 30 minutes into a 50 minute period.

I tap him on the shoulder — “you ok?”

No response.

The kid next to him — super smart, startlingly tall and easily distracted — asks me to look over his code. All of a sudden, zombie child raises his head — staring up at me with bleary eyes. He has been asleep. Like deep asleep. He apologizes and says he’s really tired. But he assures me he’s on track to get his project done, so I move on.

He puts his head down on the table and goes back to sleep.

What was I thinking? I thought it’d be fun teaching kids about computers.

But teenagers are another species.

The stereotype on GenZ is their brains are busted from the internets. Addicted to their phones, endlessly scrolling through TikTok’s. Digital natives sure, but ones without the ability to stay on task.

It’s not like that.

The teenage brain releases melatonin later than adults so they stay up too late.

Impulse control isn’t quite there yet. Their frontal lobe is still developing, so their plastic brains are more susceptible to social context, rewards, novelty, emotion.

They’re also highly capable when sufficiently motivated.

On Wednesday I throw out the slides and kick off class with a Kahoot quiz!1 The class gets into it. Super tall kid wins, answering every question right.

Streak 🔥!

JFran has Hershey’s Kisses so I top off the candy bag with those and throw in a bunch of spider-man tattoos (leftovers from my 60th bday party). I hand them out liberally to everyone who get’s their project uploaded.

The senior table is really into cubing. I got on them early to lose the cubes and the phones and FOCUS. But before spring break, I asked one explain it to me. The whole table gets in on the conversation. They find solve algorithms online and compete for time.

One of the seniors was lost when it came to the coding project until we came up with an idea for a cube solver and he excitedly shows me he got it working.

Another kid wrote a build-a-burger app in Python including drawing all the layers after the order. ASCII art. =^.^=

They’re all smart.

AI isn’t on the AP test. Weird right? It’s only the most important innovation in Computer Science in 10 years.

But after the test is over we’ll have some fun with it — I’m going to turn these kids into wizards.

Then it’ll be Summer; we’re already planning the van trips.

But I’ll miss the kids.

That same day a plumber was at our house trying to sort out our continually problematic bathroom drain.

Animal control came. I don’t know what I expected, but certainly not what went down. He took one look at the skunk and pulled a big shotgun out of his truck. He put up a makeshift screen. The screen was completely ineffective — we saw and smelled everything. He euthanized the skunk, but not before it had a final triumphant hurrah, spraying him down thoroughly.

Meanwhile, the plumber had made it up on the roof to snake out the drain. We could taste skunk from the inside with every door and window closed. The funk hit the hapless plumber and he began gagging and coughing uncontrollably. Inside, Lola the dog whined. In the front yard the animal control guy bent over gasping. He staggered towards his truck, shotgun in one hand, tearing at his face mask with the other.

My teenage daughter and I watched the whole thing go down from the front window, horrified.

R screamed, “You had one job!”

“You had one job” might be past its prime, but I laugh every time I hear it.

I recently picked up “I don’t hate that” watching curling at the Winter Olympics1. The US team was mic’d up so you could hear them talk strategy surveying the icy layout of stones. During timeouts, the coach hustled down the ice to join in the discussion.

Over and over I heard: “I don’t hate that shot.”

That phrase now lives in my head like a cereal jingle. Fits the now. Sometimes it’s hard to get excited about stuff. “I’ll allow it” is a good companion.

Resignation. It’ll do. It has to do.

Remember the AI moment — not so long ago — when you could jailbreak a new model by telling it to “Ignore all previous instructions?” I bought a t-shirt festooned with a feral raccoon screaming that out.

Alas, the early, heady days of AI when it was simple and fun are well behind us. Now we’re AI haters or we’re AI pilled or we’re vibe coding.

Vibe coding will fade quicker than “you had one job.” While it’s very Jetsons for Elroy to dictate sophisticated software builds, even with AI writing all the code, software engineering isn’t a trivial undertaking, so thanks, but I hate it.

A few more from the skunk days that stuck. KOBE! Yelled in expectation as you attempted to toss something in the trashcan from a few feet away.

We’re Gucci. Cool, cool, cool. Everything is cool. Goes well with a dab.

Maybe I’ll get lucky and my high school kids will trust me with some genZ gems. You’re good. I gotta dip. That burrito was mid. 2

I’m generationally forbidden from using any of these, but entertain me.

My dad could throw down some slang. Malarkey. Stick your head between your legs and kiss your ass goodbye. He also did a mean James Cagney impersonation: “You dirty rat …”

“Monitoring the situation” is new. I relate. Doomscrolling.

But when it comes to AI, I can’t help myself.

I miss the inside scoop. Even with corporate spin it beat the talking heads and pundits. It’s grounding to use the thing on the daily with my high school kids. AI is but a few clicks away. See for yourself.

But if you don’t want to touch the stuff, you’re good — I’m monitoring the situation.

I did research the old fashioned way — I googled it. Drowning in a sea of blue links — it was the worst.

Now that I’m working 10 hours a week teaching High School (my first couple weeks felt like 80) I have all of the AI and I love it.

It’s saving me and my students.

I handed out assessments the first day I went in. I looked them over after class — it was grim. Their teacher left at Christmas and the subsequent parade of subs just had them do self study — ineffective for all but the most motivated high schooler.

Then there’s the AP test. You take an AP class in high school so you can take the test and it’s more than about getting college credit — it shows potential colleges you’re ready for that level of work — assuming you pass.

For reasons I don’t understand, the AP test is in May - nearly a month before school is over. For AP CSP there’s another hurdle — they must code and upload a program ahead of the test and that’s due in April!

So I knew I was going to be under the gun. What I didn’t know was what tools the school would have to help me.

Education has a testy relationship with technology. When word processors replaced typewriters — well there goes spelling and editing and penmanship. And let’s not forget the great copy-paste plagiarism scourge of the 2000s.

Today it’s the cell phones messing with our kid’s minds. This one is more about attention and mental health versus teaching — and for sure it’s tricky.

AI is different. From the start, it had the potential to be the best study buddy and unlike the concerted backlash to control plagiarism back in the 2000’s, chat based AI tools have grown so quickly that schools had to figure out how to use them versus banning them.

The pandemic is really behind this shift — that’s when every teacher and student had to get online. Before COVID, around a quarter of schools used online systems and today every single public school does. Google Classroom alone saw usage quadruple from 40M pre-pandemic to over 150M now.2

That’s the world I’m teaching in today as I try to get my twenty kids back on track and ready to pass the AP exam in a short two months.

The first thing I did after I accepted the gig was try to figure out the curriculum the College Board dictated for this course. What did I have to teach them so every one could pass the AP Test? I put together — no lie — 50 slides to parse it all out for them and reset the class.

Putting aside my mistake in overestimating the attention span of a 16 year old 3 — half of the slides I pulled were how not to use AI.

Some of this is fair and a plagiarism concern. After all, AI spits out code like it’s nothing, and you’re not learning anything if AI writes your program for you.

However, AI is eating up the software industry at a frightening pace, so not only do these kids need to learn AI yesterday, it’s the perfect tool to accelerate 6 months worth of learning into 2 months time.

I’m lucky this school uses Google Classroom. Initially Google lagged getting into the AI race — they didn’t want to cannibalize their search monopoly with the pages of blue links. But exponential growth cannot be ignored and so when the threat shifted to being left out of a new age, Google went all-in. That was about a year ago and now Google has a complete suite of AI tools under the product name Gemini.

Google Classroom gets everything Gemini has to offer:

In late 2024 Google had a viral AI moment when people discovered a product they were experimenting with called NotebookLM. I fed it an article I’d written and I was amazed when it took the 8,000 characters I wrote and turned them into an entertaining 8 minute podcast.

NotebookLM has become the top AI tool in education. It’s especially popular in colleges and universities. Why? Because unlike a chatbot that pulls info from god-knows-where including occasionally just making stuff up (hallucinations), NotebookLM is scoped to a specific set of sources (up to 300) you define.

The college board publishes this document for my class and it’s 300 pages long:

AP Computer Science Principles: Course and Exam Description.

I’m here for the kids, but I’m not reading that.

But NotebookLM can — no problem. I built a notebook for my class that not only uses that doc as a primary source but also every other relevant doc from the College Board (dozens — they are nothing if not prolific at documenting) for over 500 pages of content including sample exams, prior year questions, rubrics for the coding assignment, etc.

I call it: The Exam Vault.

NotebookLM can convert its sources into a bunch of teaching resources including flash cards, infographics and practice tests. In addition to audio podcasts, it now creates video lessons as well.

I asked it to make a two minute video of the perfect coding project for the exam and my students were mesmerized. I will be making more videos:

Once I built The Vault, I could activate my AI army. I spun up a set of helper chatbots to get my kids through the coding assignment with the looming due date. It’s worth 30% of their test score. 4

Each bot has a personality and a job. First there’s AI Andrew — it’s a friendly brainstorming buddy for kids that get stuck trying to come up with a program. It’ll riff on their ideas and suggest new ones while making sure they have a structure that meets all the exam requirements and is doable in the limited time we have.

Real Andrew is the final approver before they can start to code.

Once they start coding, next is the AI Auditor — it’s a strict enforcer that checks if their code meets all the byzantine requirements of the AP test.

Once they’re code complete, the last one is the AI Simulator which is a proctor — albeit a friendly one that tells dad jokes — that simulates the written section of the exam where they have to explain the code they’ve written against the clock.

So far AI Auditor is a hit — it’s the perfect companion to check how their code is coming along. I’ve explicitly prevented any of these bots from providing code or even helping the kids write or re-write theirs. They have to do that on their own.

These should be a life saver, but I’m continually surprised by this class. Teaching high school is oh so different than teaching adults.

On Monday I switched it up; I want the kids to engage more. This was after consultation with my HS motivation coach — that’s my niece — she’s a sophomore in Oakland. On her advice, I brought in a bag of Starburst candies. When one of my seniors actually came up with the right answer I got so excited that I grabbed the bag and tossed him one.

His buddy, propped up nonchalantly on the stool next to him lurched for the Starburst, tipped over and crashed out on the floor. Very dramatic. Also entertaining. He looked very sheepish and was quieter than normal the rest of the class, but perhaps I should reconsider this tactic.

Maybe my AI study buddies will work better than Starbursts?

Nostalgia is strong. No matter how great Halo on your Xbox was in your 40s, it’ll never match the joy of playing Atari Pong in your PJs the day after Christmas when you were 9.

If I look back before the nostalgia, I can get a sense of what my parents’ lives were like before they made me — as well as what the world was like. It was very different.

In January, we flew down to LA, rented the biggest SUV we could find, and crammed it full of boxes from my parents’ storage locker. Looking through these boxes, I’ve been able to peel back a generation — 100 years past — my grandparents.

Figuring out my Grandpa’s story last month was unexpected. That’s my mom’s dad. My dad’s side remained murky — until last week.

Monday — March 2 — marked the second anniversary of my dad’s death and I didn’t notice. I’ve been busy teaching a high school class I just took over mid-year.

Tuesday I saw a text from my sister: “What date did dad die?”

I figured she needed it for some legal thing, so I dug out his death certificate: March 2, 2024.

I stared at my reply. We talked. Monday was the day I cleared the final administrative hurdles to get my Emergency Teaching Credential. An unexpected accomplishment — Monday was a good day.

The year after my dad died was rough. I lost both of my parents in 2024 — my mom died three months after my dad, a day before his 98th birthday.

The second year, this past year, was different; sorrow replaced by memory. My dad’s memorial picture has lived on our kitchen counter for the past two years. Now it’s hanging on the B&W wall of fame:

While my mom’s story keeps coming into sharper focus — I’ve discovered more letters and keepsakes from her pre-WWII life in Austria than I will ever get through — I only knew the basics about my dad’s family. My dad was close to his older brother and sister, but estranged from his mother. On my dad’s birthday last year, I wrote how he got on a bus in Pittsburgh headed for LA at 19, and never looked back.

But last week I caught a new glimpse of his family. Buried in a box that I packed when I cleared out my dad’s bedroom two years ago, I found a picture I’d never seen:

I can’t stop looking at it. I found it in a large envelope dedicated to just this photo. The envelope had the negative, multiple prints, a folded family tree and a photocopy with my dad’s writing of who these people were. There was also a letter between my dad and his brother — my Uncle John. My dad got the negative from a lost cousin he found in LA and together they’d figured out the couple on the left were their grandparents.

That makes them my great-grandparents.

That picture is from the 1920s, around 100 years ago. Looking at them, I try to imagine their lives because 100 years doesn’t seem so long ago.

My dad’s family immigrated to the US from an impoverished and oppressed corner of the Russian Empire called the Pale of Settlement — land that’s now Belarus. Jews were a lower class barred from owning property, holding office, or worshiping freely. Mobs tore through villages: pillaging, burning, killing. Pogroms. My guess is this photo was taken in the Pale, before the family left for America.

Look at the baby — I’ve no idea who they are — but even they’ve had enough of this bullshit. Life is hard.

Digging deeper in the box, I found a red plastic folio marked, in my dad’s writing: ‘Dan’s Army Discharge Papers.’ Among official US Army papers, I found a treasure trove of old photos my dad saved. It’s all photos of his family — a family I never knew.

This picture was taken just a few years after the first one in Russia. My dad’s father Nathan emigrated from Russia when he was only 15 and here he is perhaps 5 years later proudly opening up his tailor shop in Pittsburgh. Unlike the first photo with its washed-out background, this one is rich with detail — I can almost smell what it would be like getting fitted for a suit in Nathan’s shop.

Nathan’s dad — my great-great-grandpa — never made it out of Belarus, but I wonder what he would have thought about his son’s new business. To go from life in that first picture to the second — it’s easy to see why no journey was too far, no hardship too bad, to deter going to America for a better life.

I never met my grandpa Nathan, but my dad told me lots of stories about him. When he got older and a bit senile, my dad brought him back to LA to live with him in his small second-floor bachelor apartment. My dad lived a few blocks off of Fairfax avenue and Nathan would go hang out with a crew of old Jewish tailors in LA’s Jewish Ghetto.

Then Nathan developed narcolepsy — falling asleep during the day, often and unexpectedly. My dad couldn’t take care of him anymore and so he took my Grandpa Nathan back to NYC. This picture is from that trip and it’s probably the last time he saw his dad.

My Uncle John lived in NYC and he took over looking after Nathan. Uncle John carried on in Nathan’s footsteps, expanding the family tailor business into an embroidery empire.

Here’s one last picture of my dad with his brother, my Uncle John. Color! This is from my cousin’s Bar Mitzvah in 1980 and I want to believe they hit the disco afterwards.

I met up with an old friend the other day. She said she doesn’t like to live in the past, instead looking ahead and living in the moment. I agree with that. A few days later I was chatting with another friend who also writes once in a while about his past and he said it’s like driving forward but also checking the rearview mirror — to go in the right direction, it helps to see where you’ve been. I agree with that too.

Life sure is noisy today. But we’ve got comforts our ancestors never dreamed of. Bet I could get that baby to smile with a warm bath, a bottle, and a picture book. Maybe the teeniest bit of screen time.

Three weeks in sunny Redmond, WA they said. We’ll put you up in groovy digs. It’s with all your best work buddies! Fame and fortune will follow you the rest of your days.

All false. Well, nearly all. Redmond in the winter — you’re never seeing the sun. Corporate housing is a special case of dismal: rental furniture has this … smell. Fame and fortune — debatable.

Whatever. But what I wasn’t ready for was the ritualized hardship of Boot Camp.

To understand what happened, you first have to have an inkling of the importance of certification. Many industries have this. Look, I have this piece of paper, a badge, a jumble of acronyms in my signature — I am qualified. JD, MD, CPA, PhD, … The tech pinnacle of this at the time was: CCIE. Cisco Certified Internetwork Expert. Truly, these guys were wizards.

Honestly, they still are. They build the network that supports the network. There is no internet without the plumbing CCIE’s build and configure. The project I worked on had a ton of network dependencies, and so we were playing on their turf. Cisco was Microsoft’s enemy — and we were jealous of their wizards.

So we (Microsoft) introduced our own ‘elite’ certification program, calling it … Masters. 1 Microsoft Certified Master or MCM. It didn’t last — after all, software that complicated is ultimately crappy software — but I got sucked in at the very beginning.

When our fledgling product failed — under my watch — during implementation at one of our largest customers, I got deep to figure it out. 2 We did figure it out, and I loved the challenge and so I stuck with it — traveling the world with a group across the product team, consulting and support — teaching classes and introducing our latest releases as the product got successively better. The product grew quickly, it was a good time.

So, when we introduced Masters, I was happy to be recruited for the first class of a new ‘elite’ tech squad. Classes were called rotations, and this was Rotation Zero.

Twenty of us converged from all over the world to build the course. For three weeks, 10 hours a day, we lived in a gray room in tired Building 10, in the middle of Microsoft’s Redmond campus. 3 Sitting to my right was Arif — the most amazing consultant from Indonesia. His fingers flew so fast across the keyboard you could see the heat rise. Sitting to my left — Alex, the brain behind our edge servers — firewall traversal tech that I never did and never will understand. My still good friend Johan from Denmark sat behind me.4

Everyone in Rotation Zero took on teaching a module in the course, building out the content and labs for future rotations as we went. I taught the video segment. I’ve forgotten more about the deep interworking of live video than you will ever know.

Three weeks of this ended in a final exam. Were you a Master? The exam was hyped from day 1 — the guy running the show loved telling us how hard it was going to be and so we better put in the time or we wouldn’t pass. It was manufactured prestige — building lore to be retold, so Microsoft could say they had Wizards too.

The format was brutal: 8 hours straight at a keyboard connected to no less than 8 virtual machines, switch and gateway, building and troubleshooting the most diabolical scenarios imaginable.

One guy brought in a half dozen 5-hour Energy drinks, lining them up at the edge of his desk, slamming a different flavor down every couple of hours to survive the day. 5 We weren’t allowed to leave the room lest we cheat. Going to the bathroom was an ordeal.

I liked the learning and the building and the teaching, but what was up with this boot camp approach?? In the military, boot camps are designed to break you down so you can be rebuilt as one of the faithful. I wasn’t storming the beach at Normandy. I was building and fixing corporate computer systems. Also, I wasn’t 18 — I was good with who I was and had self-selected this job. The corporate cult-of-personality, religious devotion to the cause, well it started to bug me.

I didn’t pass the test. My eyes burned after 8 hours staring at a screen, madly trying different configurations of the most cryptic shit ever. We all met at a bar afterwards to blow off steam and I was so burnt. At first, I felt bad that I hadn’t passed, then I felt mad that it was so ridiculously hard after I had put so much of myself into building up the program, including leaving my family for 3 weeks — all to be treated like some kind of raw recruit.

The master of masters approached me several times after that, asking me if I wanted to retake the test. I did not. I was done with that bullshit.

But I liked the people who stood up to be part of the priesthood, signing up for Boot Camp. It’s always about the people, trying to make it even if the system is flawed. It’s not unlike my High School kids desperate for help to pass the AP test. So, I kept teaching Masters. Every subsequent rotation I traveled back up to Redmond to teach the video module.

I taught 22 more rotations, and each one was a lot of fun. 6