Chris Albon
@chrisalbon
Director of ML and Data Eng @Wikimedia.
There are only two types of evals: 1. The world's hardest academic problems, solved by humanity's most elite members in fields like mathematics and physics. 2. "Draw a duck riding a bicycle"
Investor: "Here is $50,000, with some fiscal discipline and some instant ramen meals you'll have enough runway to--" Startup:
your fancy ai startup might be monetizing agents or whatever, but has it ever been sent a commemorative seed round sword
I am still unreasonably proud of this.
I loved this video so much I vibe coded a simple playable version of it in four hours. Code and link to play is below.
I have been away from Twitter all day and I feel like I missed an episode of the show.
I've spent 4 hours last night and 4 hours tonight vibe coding. The output speaks for itself, but troublingly I don't feel like I am a better coder, despite 8 hours of staring at an IDE. Something I need to think about.
Using "claude -p 'insert your prompt here'" as a "free" (i.e. included in Claude Max) Claude API for testing and experimentation has been a game changer. Basically you can use it as a workaround to using the proper API in your code. credit for @dorkitude for pointing it out
I know I have a ton to learn about Claude Code and Agentic Programming because I've been doing a project a night for two weeks now and I cannot predict when CC will totally grok a problem and fix it and when CC is seemingly trying to play dumb
I coded a lot of my PhD work with a streamer (@CohhCarnage and @Swiftor) in a side window. I was working late at night and it made things less lonely.
talked to an engineer maybe 10 years ago who told me how he worked with his fav streamer on in the background unsurprisingly, people treat video podcasts the same. i think generationally folks dont get it but we did this with linear TV and netflix too nytimes.com/2025/07/20/sty…