-
Alibaba puts strong foot forward with new model and glasses: The Chinese e-commerce giant just dropped its latest reasoning model that’s outperforming OpenAI and Google across key math and coding benchmarks. The ‘Qwen3-235B-A22B-Thinking-2507’ has been released under the Apache 2.0 license, meaning enterprises can download, modify, and integrate it into their systems without restrictions. The company also launched its first AI-powered glasses, set to hit the market later this year.
-
Promising new paper stirs up excitement in the AI community: Researchers have unveiled ASI-ARCH, a research system that designs novel AI model architectures all on its own. It connects 3 LLM-based agents — a ‘Researcher’ that creates code, an ‘Engineer’ that trains it, and an ‘Analyst’ that spots patterns — in a self-improving cycle, outperforming models designed by humans. If true, it could overhaul AI research from a human-limited process to one that scales directly with computing power (translation: AI model progress will get a lot faster).
-
Meta taps star researcher as head of its superintelligence unit: The company has named former OpenAI researcher Shengjia Zhao as ‘Chief Scientist’ of its superintelligence effort, adding some more firepower to its all-star lineup. Zhao, who helped create ChatGPT, GPT-4, and o1, joins former Scale AI CEO Alexandr Wang in what looks like a formidable leadership duo.