Chinese model Qwen2.5, Sam Altman gone from Board, SocialAI: a social network without people — the top 3 AI news stories of the week

#1 — Chinese Qwen2.5 successfully competes with GPT-4

The new Chinese open-source model Qwen2.5 is making waves in the open-source community. The model has been released by Alibaba Cloud in three versions:

• Qwen2.5 — in sizes 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B.
• Qwen2.5-Math — in sizes 1.5B, 7B, and 32B on the way.
• Qwen2.5-Coder — in sizes 1.5B, 7B, and 72B.

Qwen2.5-Coder 7B has outperformed the initial versions of GPT-4 in certain benchmarks. Meanwhile, the basic Qwen-2.5 at 72B surpasses GPT-4o on some metrics and is catching up with Sonnet 3.5. We are looking forward to the release of Qwen2.5-Coder at 72B — this model has the potential to become the new SOTA in development!

#2 — Sam Altman gone from OpenAI Safety Board

Sam Altman has left the OpenAI Safety Board. In a press release, the company announced that the Safety and Security Committee will now be an independent Board oversight committee headed by Carnegie Mellon University professor and Director of the Machine Learning Department, Zico Kolter. There is speculation that Altman was removed following complaints from former OpenAI employees about his capricious behavior and disregard for safety rules.

#3 — An introvert’s dream: SocialAI, a social network without people

The world has gone slightly mad — and SocialAI, a Twitter-like social network without people, has emerged. On the network, users interact with agent bots. Upon registration, you can choose your preferred type of followers, ranging from nice guys to harsh critics. Agents generate numerous responses to user posts. It sounds pointless, but it’s strangely addictive. If the responsive agent bots were transferred to real social networks, it could enhance user engagement as effectively as algorithmic feeds. From an ethical perspective, however, the idea is questionable.