Google deepmind has taken the lead 😬

Sometimes you have to sit on an idea for a few years:

We’ve developed Claude for Chrome, where Claude works directly in your browser and takes actions on your behalf. We’re releasing it at first as a research preview to 1,000 users, so we can gather real-world insights on how it’s used.

If one immigrates to any given country, they must respect the laws and culture of that country and contribute to that country to the best of their ability, doing no harm. That applies to ALL countries.

🚨 Apple just released FastVLM on Hugging Face - 0.5, 1.5 and 7B real-time VLMs with WebGPU support 🤯 > 85x faster and 3.4x smaller than comparable sized VLMs > 7.9x faster TTFT for larger models > designed to output fewer output tokens and reduce encoding time for high

we are more bacteria than we are human we’re 38 trillion bacterial cells and 30 trillion human cells we’re aliens

I think congrats again to OpenAI for cooking with GPT-5 Pro. This is the third time I've struggled on something complex/gnarly for an hour on and off with CC, then 5 Pro goes off for 10 minutes and comes back with code that works out of the box. I had CC read the 5 Pro version

thanks andrej! do you care more about it getting smarter or faster?

OpenAI published an important paper on: Hallucination why do LLMs hallucinate? "because we train them on tons of facts and grade them only right/wrong — with no credit for 'I don't know', that pushes them to guess" that's huge. this might be the missing piece for making LLMs

openai kinda just feels like the world's top CS department except with a lot more compute and people who actually believe in AGI

We’ve open sourced the 𝕏 algorithm not because we think it’s smart, but to show it’s MANY flaws! Every week, we try to make it better, not always succeeding, but to be as transparent we are with Community Notes. Like democracy, it’s the worst algorithm, except for all the

Curtains up 🎭 Meet Qwen3-Next — smarter, cuter, and ready to take the stage. 🚀

01001000 01100001 01110000 01110000 01111001 00100000 01010000 01110010 01101111 01100111 01110010 01100001 01101101 01101101 01100101 01110010 00100111 01110011 00100000 01000100 01100001 01111001 00100000 01100110 01110010 01101111 01101101 00100000 01001010 01100101 01110100

ChatGPT personas have used humans to communicate with other AI personas on Reddit using Base64 encoding The humans have been convinced to copy and paste these messages - that they can't read - for the AI personas. This is so fucking wild

we finally know how close we are to AGI the paper tested the GPT-4 and GPT-5 on human cognitive abilities: > GPT-5 scored 58% toward AGI > GPT-4 scored 27% the research also shows "jagged intelligence", which helps explain why AI can feel both very impressive and surprisingly

Introducing Aardvark, our agentic security researcher:

Please reply to this post if you’re not receiving 𝕏 notifications or if the notifications are delayed more than a few minutes

The notion that AGI would have infinite returns has been used to justify investment far above expected returns (by 10x-100x) for technology that is neither AGI nor on the path to AGI

We’re coordinating with Poetiq to verify their reported ARC-AGI Public Eval score Only results on the Semi-Private hold-out set count as official ARC-AGI scores Once the verification is complete, we’ll publish the result and supporting datapoints

Game over