2 ms·
There's a reason the top labs aren't releasing their frontier models anymore, and instead keep them in-house and use them to fine-tune smaller models. Because i
by NitpickLawyer 2y ago
There's a reason the top labs aren't releasing their frontier models anymore, and instead keep them in-house and use them to fine-tune smaller models. Because it works! It's the same reason o1 doesn't give you the "thinking" steps. Distillation works. It gets you ~80% of the way, as evidenced by the qwen/llama distillations of R1.
The "walls" aren't what they appear to be.
- maeil 2y agoGemini 2.0-flash-thinking shows reasoning steps, it's only OpenAI who don't. Anthropic gives their system prompts. Looks to be SamA wanting to keep his company shrouded in mystery to appear stronger than it is.