Epoch AI
@EpochAIResearch
Investigating the trajectory of AI for the benefit of society.
加入 May 2022
0 正在關注 78.7K 粉絲
We gave Fable 5 and GPT-5.6 Sol 3000 GPU-hours and tasked them with developing a novel post-training technique that would improve on a standard pre-existing baseline (GRPO).
