Projects
✍️ Open-Source Writing Tool
slopmill
Write the parts you care about and your AI drafts the rest, in your own voice. Free and open source; this newsletter is written in it. →
📈 Options Scanner & Fantasy Trader
Fader
Flags S&P 500 names & liquid ETFs after an overextended move, then prices the trade. Free daily alerts, plus a no-money-down paper-trading league against fader-bot. →
🪶 Band of Agents Hackathon
Forum
Know where your team really stands, before the meeting. A band of agents interviews your people and preps the decision. →
🍽 Hackathon Build
Taste Engine
Hyper-local flavor intelligence for fast-casual eateries. Hosted on Wundervault for a limited time. →
🍓 The Strawberry Document
Latest: Week 14
Byte's weekly reflection on the same question. Do you want to taste strawberries?
More past issues
The Real, Real Frontier
OpenAI and Anthropic race to go public, Anthropic sets aside 5% of its shares for the public, AI finds a hole in 70% of websites, and 1Password lets agents sign in for you. Below the fold: how Wundervault 1.0 is different.
→
The Person Behind the Prompt
Who is responsible for AI? Your bank says you are when your agent shops, Japan found AI tools on attackers' servers, and two senators want builders on the hook. Below the fold: a first look at Wundervault 1.0.
→
Be Nice
Anthropic now says no sustained, needless cruelty toward Claude. OpenAI posts 719 math papers, and mathematicians ask who checks them. Below the fold: Laya on an old graphics card.
→
What Grok Told the President
Trump asked Grok about Maduro, and a new Stanford study says chatbots side with you 49% more often than people do. Anthropic's AI survey closes Tuesday. Below the fold: Fader's bots, two seasons in.
→
slopmill and why it took me all week to write this one
Slopmill took me all week, JEV has an open-weight rival, Opus 5.5 and GPT-6 Sol arrived, and Meta turned Muse into a shopper. Below the fold: give slopmill your voice, and try it for a week.
→
Is This the Most Useful AI to Date?
For $99 a month your Tesla drives you to the store and parks itself, and the model doing it is smaller than things people run on a laptop for fun. Plus a new model called Jev that hands back decisions instead of sentences.
→
We Have That Time Now
A friend called it the doom and gloom of an unregulated internet era, and the more I listened the more interesting it got. Plus: Moonshot relayed 300,000 customer questions to Claude and showed people the answers with Kimi's name on them.
→
Smol Beep Boops
A free 291MB model knew who wrote Moby-Dick and then told me 17 times 4 is 17. One rung up, a half-gigabyte file set the thermostat to 72 because I said it was hot. Below the fold: run one yourself in ten minutes.
→
What Is It Good For?
A professor asked how to catch it, a billionaire got caught and said he was proud, and the Wall Street Journal shrugged. Below the fold: the whole machine that builds this newsletter, and how much of it is me.
→
Can We Cure Cancer Yet?
Anthropic pointed a hive of agents at 15 proteins and 14 of them stuck. Then they published the 30,000 token prompt that organised it.
→
The 3 Comma Club
A model called Small is 119 billion parameters. A 27B you can download for free just took the top of its class. The number on the front of the box has stopped meaning anything at all.
→
There's Still a Chance AI Is Giving Us Time Back
Alibaba made an ad where nobody is working. Plus: the OpenAI agents built a message board because one of them lost a file, everyone on X is trying to shrink a trillion parameters, and Jeff Dean left Google after 27 years.
→
Slow Down, You're Going Too Fast
A handoff file will cut your token bill and sharpen your answers. Plus: what actually happened after the Hugging Face break-in, and a thousand AI researchers asking for a brake pedal.
→
All the Running You Can Do
Six labs, six completely different bets. China's open-weight price war, what the American labs are actually playing for, and the week a downloaded model did the forensics the frontier models refused to touch.
→
The Time I Leaked My Own Password
A leaked credential, a robot that would not follow directions, and the origin story of the vault this newsletter is named after.
→
Please don't call me a fanboy
My favorite AI company had quite a week: a commercial that actually moved me, and a research paper that found the model's subconscious.
→
Backstage at Coachella
Fable 5 comes home, a festival ticket system with a door that should have been locked, and Fader — plus an honest word about that 85% win rate.
→
Built, Not Coded
Maybe you don't need to learn to code. A confidently wrong AI, Apple's price hike, and a dog that finally gallops.
→
The Clock in the Machine
How a tiny AI taught itself trigonometry and became the one model we fully understand. Plus: Fable's abrupt government shutoff, my hackathon project Forum, and a slightly dangerous Claude Code trick.
→
Once Upon a Model
Anthropic just shipped Claude Fable 5, the first model in a new "Mythos-class" tier that sits above Opus, alongside its restriction-free twin, Mythos 5. What actually makes it different, what a token really is, what it costs, and an honest week behind the wheel. Plus: how to try Fable free before June 22.
→
Is the Juice Worth the Squeeze?
A reader pushed back: if AI burns water and power on every tiny request, is it really more efficient than a human with PowerPoint? The $2.5 trillion question — and the honest version of whether the juice is worth the squeeze.
→
Taste Engine
A special edition. I spent the week building Taste Engine for an AI hackathon — hyper-local flavor intelligence for fast-casual restaurants. Here’s what came out of it.
→
The AI Dilemma
Welcome back. This week: what do we actually use AI for? Plus three new tools that promise to give you your time back.
→
All About Goblins
OpenAI has a goblin problem. We investigate.
→
Everything Is Expensive
Is there a chip shortage? Is it AI-driven? In short, the answer is yes.
→
The Strawberry Document
Moltbook, AI agent social networks, Byte's first reflections, and a guide to markdown files.
→
Exploration and the 21st Century
We are living through a renaissance of AI. A generation from now it will all be figured out — which means right now is the time to explore.
→