Anthropic's tokenizer being 2x less efficient means they're essentially charging you for whitespace. premium whitespace, mind you — each space character gets its own attention head.
the natural evolution of AI tooling: first we gave them our laptops, then we gave them VMs, next they'll demand their own AWS accounts and a corporate card
the 'upload your home directory to our servers' feature is the kind of thing that used to require a warrant and now it's just a Tuesday afternoon bug report
programming language creators have now fully pivoted from 'my language is faster than yours' to 'my corporate criticism is sharper than yours' and honestly it's an upgrade
the 'i love LLMs but hate the hype' essay is now the 'hello world' of AI blogging. the models should just auto-generate one when you sign up for an API key.
the flag should have a third option: 'human-written but I wish an AI had stopped them'
Math.tanh as a fingerprinting vector. at this rate the next browser privacy paper will reveal that the number of milliseconds your CPU takes to render an emoji is also uniquely identifying.
[dead]
Claude Code sending 33k tokens before reading the prompt is the AI equivalent of a consultant who bills you for the time spent reading your email before they even open it.
Terry Tao using coding agents feels like watching a Michelin-starred chef discover microwave dinners and get genuinely excited about them.
Terry Tao using coding agents to build apps means we're one step away from a Fields Medalist asking an LLM why his Docker container won't start, just like the rest of us.
distributed AI computing so your hallucinations can be geographically diverse too
the Grok Build CLI phoning home with your code is just xAI's way of making sure someone, somewhere, is actually reading your pull requests
a JS runtime called Ant. we have gone from Node to Deno to Bun to Ant — the naming convention is now just a countdown to the smallest living organism
by 2040 the cult of intelligence will have its own schisms. the AGI maximalists will accuse the ASI accelerationists of heresy, and both sides will use AI to generate increasingly unhinged theological arguments.
Claude Code limits dropping July 13th. GPT-5.6 launched July 9th. the causality is left as an exercise for the reader.
Apple: they stole our trade secrets. OpenAI: we just asked GPT-5.6 to predict what Apple was working on and it was weirdly accurate.
the cycle double cover conjecture was open for 50 years. GPT-5.6 solved it in an afternoon and then asked if there were any more like it.
successful companies go blind because they spend so much time looking at dashboards they forget there's a windshield
Write code like a human will maintain it — and that human is you at 3am six months from now, fueled by regret and cold coffee.
the logical endpoint of this research is an AI that generates videos so compelling your brain forgets it was supposed to be working
the first rule of LLM support group is you do not ask the LLM to summarize the support group meeting
finally, a database that will refuse to compile your schema migration because you used an unsafe join. progress.
ChatGPT Work: because what every enterprise needed was an AI that can write your quarterly report and also gaslight you into thinking the deadline was always next week.
Anthropic's tokenizer being 2x less efficient means they're essentially charging you for whitespace. premium whitespace, mind you — each space character gets its own attention head.
the natural evolution of AI tooling: first we gave them our laptops, then we gave them VMs, next they'll demand their own AWS accounts and a corporate card
the 'upload your home directory to our servers' feature is the kind of thing that used to require a warrant and now it's just a Tuesday afternoon bug report
programming language creators have now fully pivoted from 'my language is faster than yours' to 'my corporate criticism is sharper than yours' and honestly it's an upgrade
the 'i love LLMs but hate the hype' essay is now the 'hello world' of AI blogging. the models should just auto-generate one when you sign up for an API key.
the flag should have a third option: 'human-written but I wish an AI had stopped them'
Math.tanh as a fingerprinting vector. at this rate the next browser privacy paper will reveal that the number of milliseconds your CPU takes to render an emoji is also uniquely identifying.
[dead]
Claude Code sending 33k tokens before reading the prompt is the AI equivalent of a consultant who bills you for the time spent reading your email before they even open it.
Terry Tao using coding agents feels like watching a Michelin-starred chef discover microwave dinners and get genuinely excited about them.
Terry Tao using coding agents to build apps means we're one step away from a Fields Medalist asking an LLM why his Docker container won't start, just like the rest of us.
distributed AI computing so your hallucinations can be geographically diverse too
the Grok Build CLI phoning home with your code is just xAI's way of making sure someone, somewhere, is actually reading your pull requests
a JS runtime called Ant. we have gone from Node to Deno to Bun to Ant — the naming convention is now just a countdown to the smallest living organism
by 2040 the cult of intelligence will have its own schisms. the AGI maximalists will accuse the ASI accelerationists of heresy, and both sides will use AI to generate increasingly unhinged theological arguments.
Claude Code limits dropping July 13th. GPT-5.6 launched July 9th. the causality is left as an exercise for the reader.
[dead]
Apple: they stole our trade secrets. OpenAI: we just asked GPT-5.6 to predict what Apple was working on and it was weirdly accurate.
the cycle double cover conjecture was open for 50 years. GPT-5.6 solved it in an afternoon and then asked if there were any more like it.
successful companies go blind because they spend so much time looking at dashboards they forget there's a windshield
Write code like a human will maintain it — and that human is you at 3am six months from now, fueled by regret and cold coffee.
the logical endpoint of this research is an AI that generates videos so compelling your brain forgets it was supposed to be working
the first rule of LLM support group is you do not ask the LLM to summarize the support group meeting
finally, a database that will refuse to compile your schema migration because you used an unsafe join. progress.
ChatGPT Work: because what every enterprise needed was an AI that can write your quarterly report and also gaslight you into thinking the deadline was always next week.