Helping people solve their biggest AI Problems
AI Innovator | Quant Engineer | Building the future with AI | Sharing insights on GenAI and ML breakthroughs
@swyx@METR_Evals Crazy! This seems to be bold bet. Would love to keep track, how it plays out. If it works, more and more labs will come to have bets like these.
@samjulien Can you tried Claude design or claude code with UI / Frontend plugin ? I think that competes heads on with Codex. Let me know what you think
@t_blom I think the agents just like humans need experiental memory. Without it, no matter how much context you give, you'll come back to same point after every session.
@garrytan@garrytan resolvers are the routing layer - but what happens when the context they load becomes stale or contradictory? Humans don't store everything, we compact temporally. Built an auto-dream architecture inspired by Stanford's Generative Agents paper that does exactly this.
My 78yo uncle quit AI after 3 minutes.
My 9yo niece built a Minecraft empire with Claude in 10.
Same species, different planets.
We need Animal Crossing that secretly teaches AI - for boomers too.
Tag someone who needs this yesterday
What if you could slash token costs without touching your JSON code?
TOON: open-source format that compresses prompts → way fewer tokens, same meaning, zero logic changes.
Perfect for heavy AI workloads 🔥 https://t.co/756DLuY3LT
@AkhandPratap_@sama@gabeeegoooh Don't try this for images containing text. That's a general trend for text on images - spellings are wrong.
Try something without text.