Re:DeepSeek Harness
by alansaber on Thursday August 13, 2026 @12:58PM
(#49288800)
Agreed that it is a killer feature. US models obfuscate the COT (to A. make it look better and B. combat distillation) but > and the raw trace is fairly hard to reason about > but I still think this kind of feature is a big step in the right direction.
[ Parent
| on HN ]
Re:DeepSeek Harness
by gosolozero on Thursday August 13, 2026 @03:32PM
(#49290845)
I agree it’s a great step. But the deepseek models also don’t perform to the same level of fable/sol. If we optimize/finetune to deepseek traces, wouldn’t it be suboptimal? What would the benefit be?
You let the smarter model explore the traces and figure out where the current harness' bottlenecks are for the current LLM. Then you can adjust prompts or tools to fix those.
Can you explain how COT is essential to distillation and why obfuscating it helps defend against distillation?
Or point me in the right direction in terms of what to read.
Because they are an integral part of the solution generation.
By distilling, you are training a model via Reinforcement Learning (RL) to mimic the answer of a bigger model. To do that, you need all the steps that contributed to generating an answer.
To give you a better idea, imagine teaching a student how to solve math problem:
1. You give it the problem and the answer only (no thinking trace) 2. You give it the problem, the intermediary steps and the answer (full transcript)
I think you can agree that the second method is more likely to give a well-informed student.
Isn't distillation done through SFT by having the student model approximate the teacher's output distribution?
This might be helpful https://arxiv.org/abs/2608.09867