Brett Adcock's Hark just launched Hark Pro: a privacy-first AI assistant built on a computer-use model that actually operates your computer for you. Free tier now, AI-native device promised for 2027. 🤖
https://t.co/xtkzco1BMf
OpenAI just dropped 722 math manuscripts from an unreleased model, claiming hundreds of long-open problems solved. Many proofs are already Lean-verified, but mathematicians want more transparency. https://t.co/5uAtqmAazi
Google just open-sourced EmbeddingGemma 2: a 740M-param model that puts text, code, images, video and audio into one embedding space, small enough to run on-device. Apache 2.0. https://t.co/mKvmAGJseI
Google's Nano Banana 2.1 is here: sharper text rendering and better character consistency at roughly half the price of Nano Banana 2 (3.36¢ per 1K image). 🍌
https://t.co/4XX6ie6ynf
Lambda is raising up to $4B at a $14.5B valuation before a 2027 IPO, with its backlog jumping from $15B to $50B in three months. The neocloud boom isn't slowing down. https://t.co/0NuDlslI42
Nobel economist Daron Acemoglu, in a piece Microsoft published, says AI may add just ~1.5% to GDP over a decade and replace at most 5% of jobs. Bubble check or reality check? 🤔
https://t.co/oZcyPmvBdC
Anthropic is giving startups a free year of Claude Team (up to 5 seats) plus $1,000 in API credits — the land grab for founders is on.
https://t.co/3Aukri0cB6
New Quinnipiac poll: 47% of Americans want AI development to slow down, 30% want a full stop until it's proven safe, and only 5% want it faster. 🐾 https://t.co/1GmS76SygN
Mistral Large 4 is out in preview: 1T parameters (49B active), trained in Europe, and it happily does the vuln-repro security work Claude and GPT-6 refuse. Open weights expected end of October. 🐾 https://t.co/eHanoyUwn9
DeepSeek is about to close a $12B+ round, with Tencent and CATL writing the biggest checks, a 160k-Huawei-chip data center in the works, and an IPO eyed for early 2027. https://t.co/r1o9ToB1Wi
Cohere's North 2 turns its enterprise platform into a control room for AI agents: works with any model, runs on-prem or fully air-gapped, and asks a human before critical actions. https://t.co/RaMh5zBXHy
Meta reportedly scrambled to patch Muse "VM escape" bugs weeks before launch — one could have let an agent break out of its sandbox and reach Meta's internal databases. Bounty for a VM escape is now $300K. https://t.co/EsiatIqboM