Pinned
We've raised $7M to help companies build AI agents that actually learn and work.
@Osmosis_AI is a platform for companies to fine-tune models that outperform foundation models with reinforcement learning.
Better, faster, and cheaper.
00:00
Kasey Zhang
9,956 posts
Kasey Zhang
@_WEEXIAO
🇺🇸 | buildin @osmosis_ai; yc w25 | prev: @audacioushq @challengers_up (acq) | aspiring hedgehog; chopping wood and carrying water
- Don't use structured output mode for reasoning tasks. We’re open sourcing Osmosis-Structure-0.6B: an extremely small model that can turn any unstructured data into any format (e.g. JSON schema). Use it with any model - download and blog below!
GIF - It’s easy to fine-tune small models w/ RL to outperform foundation models on vertical tasks. We’re open sourcing Osmosis-Apply-1.7B: a small model that merges code (similar to Cursor’s instant apply) better than foundation models. Links to download and try out the model below!
GIF - riot game designers when players start building moonstone on every champ
00:00 - Replying to @theneedledrop and @LilNasXYOU TAKE THE MOON AND YOU TAKE THE SUN
- >see new codegen model >ask company if model is fully custom or just post-trained on chinese oss >they don't answer >explain the differences between fully custom and just post-training >they laugh and say "it's a good codegen model sir" >try the model >it's chinese
- Had an great time at @ycombinator HQ running RL IRL - YC was at max capacity! We wanted to focus on *applied* RL: how any company could use RL to improve agents, not just labs (i.e. what we do at @Osmosis_AI). Thank you to all the attendees, as well as our amazing co-hosts!
- Replying to @hamptonismmen used to go to war now they're fighting for euro lmm deals
