The powerful and inexpensive AI model of DeepSeek has involved the world of Tech into the storm. Altman said that Openai would accept one of Deepseek’s popular approaches.
If a rival succeeds with another approach, you may pay to change the course.
According to the Friday Reddit AMA session, Sam Altman said that Openai will do it.
The discussion mentioned some AI topics, especially Altman asked about DeepSeek. This is a relatively inexpensive top performance AI model and then involves the high -tech world in the storm.
One Reddit user asked if Openai could display “all thought token”. This refers to the idea that a new “reasoning” AI model is used to split tasks into smaller steps.
The O1 and O3 models of OPENAI use this inference approach, but only display the final answer instead of the user of the intermediate thinking.
Deepseek’s inference models, such as providing R1, display all steps to users. When Business Insider demonstrated Deepseek in a Chinese lab Deephink setting, it shared about 16 pages of mathematical procedures and provided the correct answer to severe questions.
On Friday, Altman said that the opening would follow the Deep Shek approach. “Yeah, we will show you a very helpful version of this. R1 credit that updated us.”
Open source and open weight
Meta’s chief AI scientist Yann Lecun states that the biggest point from DeepSeek’s success is the value of the open source AI model and its own model.
META’s LLAMA model is almost open and can be accessed for free details such as Weights and Parameters. By sharing such a model, other developers and many companies can customize these models for their own use.
Despite its name, Openai has adopted a more closed approach to AI development so far. Most of the models are unique and have a startup fee for access.
During Redditama on Friday Altman was asked if Openai would release some of the model weights and to consider some research.
“Yes, we are discussing. We are on the wrong side of history, so we need to know different open source strategies. Openai share this view. Not to say, it is not our current top priority, “said Altman.
Are you working in a meta? Contact this reporter from pdixit@insider.com or Pranavdixit@protonmail.com non -occupational e -mail and device. You can also safely reach him via the +1408-905-9124 signal. Your identity will be protected.
