Sam Altman说openai 有一方面也要学deep seek

[billwanhua]

Sam Altman says OpenAI will embrace two new AI approaches, one from DeepSeek and another from Meta

When rivals take a different approach and succeed, it sometimes pays to change course

africa.businessinsider.com

The CEO said OpenAI has "been on the wrong side of history" when it comes to model weights.

OpenAI's o1 and o3 models use this reasoning approach, however they don't show any of the intermediate thinking steps to users, and instead just show the final answer.

DeepSeek's reasoning models, such as its R1 offering, show every step to users. When Business Insider demoed DeepSeek with the Chinese lab's DeepThink setting, it shared about 16 pages of mathematical steps before providing the correct answer to a tough question.

On Friday, Altman said OpenAI would follow DeepSeek's approach. "Yeah we are gonna show a much more helpful and detailed version of this, soon. Credit to R1 for updating us," he wrote.

推理模型的中间步骤，虽然很多错误，但是却有很多启发性，因为如果机器不给出来，问者可能去探索那些错误的可能行。

这一点我个人使用特别喜欢，而且看网上很多人评论都说喜欢，我前几天在这里也指出了这一点。