- The powerful and inexpensive AI models of Deepseek have stormed the world of technology.
- Altman said Openai will adopt one of the popular approaches to Deepseek.
- The CEO said Openai had “been on the wrong side of history” with regard to the weight model.
When the rivals adopt a different approach and succeed, it is sometimes advantageous to change course.
This is what Sam Altman said that Openai would do, according to a Reddit Ama session on Friday.
The discussion approached several AI subjects, but in particular, Altman was questioned on Deepseek, who stormed the world of technology after having deployed the most efficient AI models which are relatively cheap to use .
A Reddit user asked if Openai could show “all reflection tokens”. This refers to the chain of thought that new “reasoning” models use to divide the tasks into smaller steps – similar to the way humans think through complex challenges.
OPNAI’s O1 and O3 models use this reasoning approach, but they do not show any of the intermediate stages of users, and rather show the final response.
Deepseek’s reasoning models, such as its R1 offer, show each user stage. When Business Insider demonstrated Deepseek with the Deepthink framework of the Chinese laboratory, he shared around 16 pages of mathematical steps before providing the right answer to a difficult question.
Altman said on Friday that Optai followed the approach of Deepseek. “Yeah, we will show a much more useful and detailed version of this, soon. Credit at R1 for updating us,” he wrote.
Open-source and open weight
Meta-Chief AI scientist, Yann Lecun, said the biggest point of Deepseek’s success was the value of open source AI models versus owners.
Meta’s Llama models are mainly open source, allowing anyone to access important details free of charge such as weights and settings. The sharing of interior models of models like this allows other developers and many companies to personalize these models for their own use.
Despite its name, Openai has adopted a more closed approach to the development of AI so far. Most of its models are owners and starting costs for access.
During the AMA Reddit on Friday, Altman was asked if Openai would plan to release some of his model weights and publish research.
“Yes, we discuss. I personally think that we have been on the wrong side of history here and that we must find a different open source strategy; everyone in Openai shares this opinion, and it is not either Our current priority, “said Altman replied.
Do you work at Meta? Contact this journalist from an email and a non-work apparatus to [email protected] Or [email protected]. You can also reach it safely via the signal to + 1408-905-9124. Your identity will be protected.