OpenAI posts 722 math papers, but verification is still partial
9 sourcesThe collection contains work from an unreleased internal model, with Lean formalizations available for many results.
New publication notifications
AI News
Models, agents, research and regulation in artificial intelligence: every story cross-references multiple sources and links back to earlier developments covered on AlexTech.ai.
Google brings search across audio and video onto the phone
6 sourcesEmbeddingGemma 2 maps text, images, audio and video with a model containing 740 million parameters.
Underdog runs AI on your computer and plans to earn from payments
7 sourcesIts invite-only beta works on Mac and Windows, keeps the model on-device and starts out free.
Opus 5.5 beats human experts at experimental research
8 sourcesTasteVal tests 20 models on eight tasks, measuring the compute needed to reach the human baseline.
Claude can edit Google files without leaving Docs, Sheets or Slides
7 sourcesAnthropic's new Google Workspace beta allows Claude to write formulas, build pivot tables, and rewrite text via a sidebar.
Nano Banana 2.1 improves image editing and keeps subjects consistent
6 sourcesThe Gemini 3 family model is also rolling out in AI Studio, Flow and Google Ads.
NASA is building humanoids for the Moon's riskiest jobs
5 sourcesA 16-person team in Houston is testing hardware and software for robots that can act without constant remote control.
Mistral previews a trillion-parameter AI model
8 sourcesThe preview is available through an API, while downloadable model weights are expected at the end of October.
Tencent opens up Octop, an AI assistant you host yourself
8 sourcesThe open-source platform brings chat, agents, automation and messaging integrations into one self-hosted service.
Codex lets a second agent review risky actions for free
8 sourcesThe reviewer can be enabled in settings to assess network requests, external tools and actions with real-world effects.
Sony Music asks platforms to remove 260,000 AI deepfake songs
7 sourcesThe tally has nearly doubled since March; the tracks imitate artists including Adele, Queen and Harry Styles.
South Korea proposes $3.49 billion for homegrown AI
3 sourcesThe plan needs parliamentary approval and will open the competition to startups and current contenders.
Info
AI-Generated Content Notice: The reports, articles, headlines, and images published in this section are produced with the aid of artificial intelligence tools and are subject to human supervision. Despite reviews and checks, content may contain errors, omissions, or missing updates. Materials are for informational purposes only and do not replace official sources or professional advice.
Follow our channels