THANKS GREG HI KSALFJASKLFJALKS

Understanding and preventing misalignment generalization Recent work has shown that a language model trained to produce insecure computer code can become broadly “misaligned.” This surprising effect is called “emergent misalignment.” We studied why this happens. Through this

1/9: Dense SAE Latents Are Features💡, Not Bugs🐛❌! In our new paper, we examine dense (ie. very frequently occuring) SAE latents. We find that dense latents are structured and meaningful, representing truly dense model signals.🧵

THATS MY GOAT HOLY SHIT

i did not realize you can't use a passport card for international travel 💀

I’m with geohot on this one :)

do not build Infinite Jest (V), do not build the infinite AI TikTok slop machine, do not build the P-zombie AI boy/girlfriend, do not build the child-eating short-form video blackhole, do not build the human-feedback-optimized diffusion transformer porn generator. save yourselves

I’m disappointed I have so many friends in sf and none of them went to the robot fight club. :(

POV: AI researchers playing poker for $50 buy-ins

guess it’s time to reveal the end goal of Transluce investigator agents is designing our clothing line

Huge thanks to Sarah Schwettmann for a fascinating keynote on "AI Investigators for Understanding AI Systems" 🤖 Sarah Schwettmann Transluce

Don’t overfit models to benchmarks. Don’t overfit benchmarks to models.

I love GPUs

It’s just entirely over

“MLPs are vector databases”

I agree

Went outside with Franklin Wang :)

Chat it’s over I’m turning 20 in 365 days

“that one roblox boat event” 🫡🫡🫡

went for a quick ride