48 hours in AI: Images 2.5, Meta Muse and an OpenAI math proof dispute
A roundup of three stories: Images 2.5 has already run 3 billion images, Meta launched a personal agent called Muse, and an OpenAI internal model solving a 90-year-old math problem sparked a dispute.
The last 48 hours in ai were unreal. >openai dropped chatgpt images 2.5. faster, sharper, smarter, with real editing tools this time. they say over 3 billion images have already run through it. the "make me a picture" era just jumped again. >meta launched muse, not a chatbot, a personal agent that lives on its own machine in meta's cloud, works in the background, and hits the glasses next. built by alexandr wang's team. the billions zuck spent finally shipping. >then openai said an internal model, "significantly more capable than gpt-6," cracked a 90 year old math problem. 10,000 agents, 88 hours. >and the nyu mathematician working on it went public. says openai caught wind of his research, raced him to the finish, and leaned on his anthropic co-author to come off the paper. openai's answer: we "can't rule out" that de-identified user data helped. >Pictures, agents, and math proofs in one week, the machines stopped just answering our questions. now they're doing the work, and fighting us for the credit. That's the three big ones Images 2.5, Muse, and the OpenAI math fight in a single arc.