GPT-5.5 becomes first to ace all 200 binary rewriting benchmarks
Recursive, founded by former DeepMind chief scientist Tim Rocktäschel, wants AI to run its own experiments. The startup just raised $650M at a $4.65B valuation.
AI
Most Recent
TabPFN-3 processes a million rows in a single pass — no tuning needed
Recursive, founded by former DeepMind chief scientist Tim Rocktäschel, wants AI to run its own experiments. The startup just raised $650M at a $4.65B valuation.
Former Alibaba AI lead starts new lab with $2 billion valuation target
Lin Junyang, the creator of Alibaba's Qwen AI models, is raising funding for a new lab at a $2 billion valuation. The startup has no product yet, but investors are betting on his track record.
MiniMax renames desktop agent Mavis, adds collaborative AI teams
MiniMax rebrands its desktop AI agent as Mavis, adding a multi-agent collaboration feature called Agent Teams that splits complex tasks among specialized agents.
Kaiming He's ELF: The language diffusion model finally works
MIT Kaiming He's team releases ELF, a language diffusion model that generates text in continuous embedding space, achieving competitive results with far less training data than autoregressive models.
Musk boasts open-source X algorithm, Community Notes fact-checks: never updated since launch
Elon Musk reiterated his commitment to open-sourcing X's algorithm, but Community Notes pointed out that the code has not been updated since its initial release and is missing critical files.
Anthropic's 'Mythos' AI Forces New US-China Talks; Huawei Chips Are Beijing's Ace
A new AI safety dialogue between the US and China is catalyzed by Anthropic's autonomous hacking model Mythos, while DeepSeek's Huawei chip compatibility strengthens Beijing's negotiating position.
‘Are You Trustworthy?’: Sam Altman Admits to Lying in OpenAI Trial
During cross-examination in Elon Musk’s lawsuit against OpenAI, Sam Altman admitted to lying in the past and faced tough questions about his financial conflicts of interest.
Amazon employees are 'tokenmaxxing' to pad AI usage stats
Amazon employees are using an internal AI agent tool called MeshClaw to inflate their token consumption on company leaderboards, gaming metrics meant to justify the tech giant's massive AI spending.
Anthropic launches Claude for Legal with 12 plugins and MCP connectors
Anthropic released Claude for Legal, an open-source GitHub repository with 12 plugins and over 20 MCP connectors for legal professionals, designed to customize AI outputs to each firm's style.