Is deepseek’s r1 model distillated from ChatGPT?

来源: QualityWithoutName 于 2025-01-30 10:34:02 [] [博客] [旧帖] [给我悄悄话] 阅读数 : (4170 bytes)

Yes, there is evidence suggesting that DeepSeek’s R1 model was developed using a technique called “distillation� from OpenAI’s models. Distillation involves training a new model by leveraging the outputs of a pre-existing model, effectively transferring knowledge from the original to the new model. OpenAI has indicated that it found evidence linking DeepSeek to the use of distillation—a common technique developers use to train models by leveraging the outputs of existing ones. 

 

Additionally, discussions within the AI community have raised concerns about DeepSeek’s methods. For instance, a thread on the OpenAI Developer Forum titled “Looks like Deep Seek R1/V3 was distilled from GPT-4/3.5 - Can anyone confirm?� delves into this topic. 

 

Therefore, it appears that DeepSeek’s R1 model was indeed developed through distillation from OpenAI’s models.

 

所有跟帖： 

• Answer from ChatGPT 4o -QualityWithoutName- ♂ 给 QualityWithoutName 发送悄悄话 QualityWithoutName 的博客首页 (0 bytes) () 01/30/2025 postreply 10:35:14

• 他们为什么不自己蒸馏？还是技不如人 -sxyz- ♀ 给 sxyz 发送悄悄话 (0 bytes) () 01/30/2025 postreply 10:49:09

• 蒸馏技 -Bob007- ♂ 给 Bob007 发送悄悄话 (0 bytes) () 01/30/2025 postreply 11:19:43

• 我刚刚问了ChatGPT 同样的问题，回答正相反。奇怪！ -方外居士- ♂ 给 方外居士 发送悄悄话 (5444 bytes) () 01/30/2025 postreply 10:50:31

• Can anyone confirm？ -监考老师- ♂ 给 监考老师 发送悄悄话 监考老师 的博客首页 (0 bytes) () 01/30/2025 postreply 11:00:55

• 上面附上了截屏。 -QualityWithoutName- ♂ 给 QualityWithoutName 发送悄悄话 QualityWithoutName 的博客首页 (0 bytes) () 01/30/2025 postreply 11:21:00

请您先登陆，再发跟帖！