Loading activity…
@neal_wu
@thinkymachines, prev new stealth co, @cognition, @tryramp, @GoogleBrain, competitive programming
AI
In its SFT tests, Tinker says Dimuon often reached lower loss in fewer steps than AdamW.
AI
Tinker, the post-training API by Thinking Machines Lab, offers a recipe using RL for calibrated event predictions.
AI
Compact multimodal model matches original performance at quarter the size.
