Deep Chic unveiled the most recent AI model specialized in mathematics with none official announcement. This appears to be intended to envision the completion of technology before the launch of the following generation model and to get feedback from the event community.
Deep chic is 30 days (local time) HubThrough Mathematics-specific AI model ‘Prouver-V2’ was released as an open source.
At some point after Alibaba announced ‘QWen 3’, there isn’t any official announcement yet. It will probably be seen as ‘stealth launch’ in the extreme AI competition.
There was no disclosure of a separate technical document. In line with the introduction of Hugging Face, ‘Deep Chic-V3’ with 67.1 billion parameters relies.
He also adopted a MOE architecture for efficient operation, and explained that it’s going to reduce the calculation burden through the use of a smaller 7B model to handle the proof process.
As well as, only the performance comparison with its own model, akin to the present V3, has been released, and there isn’t any benchmark result with the open AI or third -party models.
That is the sequel to the Pruber-V1.5, which was released in August last yr, and is a domain-specific model aiming to unravel mathematics problems. Deep Chic said in a report that he strengthened his mathematical reasoning and proof of theorem.
Additionally it is noteworthy that it was released ahead of the launch of the next-generation reasoning model ‘Deep Chic-R2’. Deep Chic has recently been quiet, but has continued to disclose the most recent research results, showing the outline of R2.
Last month, the V3 model update was announced, including improved reasoning ability, coding optimization, and improved Chinese sentiment. That is accepted as a trailer for ‘Deep Chic-V4’.
Alibaba, alternatively, emphasized that the performance is best than the ‘O1’ of the Q-One 3-ga ‘Deep Chic-R1’ and Open AI.
By Park Chan, reporter cpark@aitimes.com
