Michal Wolski @michalwols

CV / multimodal eng prev: video gen @character_ai, founder at biteai (acq by MFP), 1st employee @clairifai, CS @columbia

 Greenpoint, NY
michal.io
Joined November 2011
Replying to @natolambert
Aged like wine
1
6
Replying to @francoisfleuret
@Kantrowitz knew what he was doing putting him in that orange room
2
Replying to @wightmanr
1
GIF
Replying to @ClementDelangue
Congrats! PS: Did you castaway on some remote miami beach?
1
GIF
Replying to @ZaidJilani
Cuomo doesn't discriminate
1
3
Replying to @RadishHarmers
2
Replying to @anyscalecompute @astral_sh
Will this also support script mode, so that all deps can be define in main.py like:
1
Replying to @_clashluke @cloneofsimo
Even better: michalwols.github.io/yann/hy…
3
Replying to @Teknium
They've been very open about this for a while now, 90% of Trump's cabinet is founders fund / palantir adjacent. This is the kind of ads they're running in the middle of football games: piped.video/watch?v=UiiqiaUB…

The Future of Warfare | Palantir

Battles are won before they begin.Army vs Navy Commercial 2024https://www.palantir.com/platforms/gotham/

youtube.com
1
1
Replying to @michalwols @osanseviero
Actual winner: The man child trio of Elon, Zuck and Sama having a data center measuring contest
Great ML podcast from @andrey_kurenkov that not enough people seem to listen to piped.video/i6JECxeu6mc?si=8NmZ…

Last Week in AI #198 - DeepSeek R1, Qwen 2.5, OpenAI Agents

Our 198th episode with a summary and discussion of last week's big AI news!Recorded on 01/31/2024Join our brand new Discord here! https://discord.gg/nTyezGSK...

youtube.com
Replying to @srush_nlp
Even just looking at V3 vs R1, most of the gains are in tasks they explicitly optimized for. There are similar task specific applications of GRPO now that don't seem to show much in terms of gains even across STEM tasks
Replying to @arankomatsuzaki
1
Load more