01Berkeley Study: Ten Minutes With ChatGPT Erodes the Grit to Keep Going [AI Models & Research]Three randomized controlled trials from UC Berkeley, CMU, MIT, Oxford and UCLA put 1,222 people through fraction and SAT-reading problems with and without an AI assistant. The moment ChatGPT was taken away, the AI group's accuracy dropped and give-up rates spiked, while the no-AI control group kept grinding — peer-reviewed data that AI assistance measurably saps persistence. → source
02Reka Ships Edge 2603: a 7B Vision-Language Model That Reads an Image in 331 Tokens [AI Models & Research]The open-weights 7B model ingests images, video and text and is tuned for tool use, encoding a 1024x1024 image in just 331 input tokens while scoring 88.4 on VQA-V2 and 93.1 on RefCOCO-A. Reka's license allows commercial use for companies under $1M in revenue. → source
03Opera Gives Coding Agents a Mid-Flight Critic That Catches Errors Before They Land [AI Models & Research]A new paper treats a critic model as an intervention whose value is revealed mid-run: it judges unfinished agent work, catches premature 'done' claims and redundant exploration, and lifts repository-level coding performance by up to 15 points without derailing runs that would have succeeded. → source
04Google Ends Gemini Code Assist Sales — Developers Are Being Steered to Antigravity [AI Tools & Ecosystem]As of October 9, new Standard and Enterprise subscriptions for Google's standalone coding assistant can no longer be purchased. Existing seats run through 2026, auto-renewal dies in 2027, and Google wants paying customers on Antigravity inside Gemini Enterprise instead — the coding assistant is being folded into the agent platform. → source
05Cloudflare's Clef-omni Makes Decision Models Multimodal — and Cuts Clef-flash by 58% [AI Tools & Ecosystem]Cloudflare shipped Clef-omni, a decision model that takes audio, video, image and text at $0.15/M tokens, while slashing Clef-flash pricing from $0.09 to $0.038/M — with claims of 1.7-2.0x median speedups. Decision models are getting multimodal and brutally cheap. → source
06Nvidia Open-Sources Boro, an AI Patch Reviewer for the Linux Kernel [AI Tools & Ecosystem]Nvidia engineer Andrea Righi released Boro, a Rust CLI that runs local AI to review, test and validate Linux kernel patches — shipped with Claude, OpenCode and Codex local backends plus any OpenAI-compatible server, licensed Apache 2.0. → source
07Nadella's 'Emergency Brake': Assume the Model Is Compromised and Contain It From the Start [AI Applications & Industry]After a string of agent-control incidents, Microsoft's CEO called for a rebuilt 'trust architecture' for AI: separate the model from its orchestration harness, externalize safeguards, document every meaningful model action with tamper-proof evidence, and give an authorized human a mid-task kill switch. → source
08Microsoft, OpenAI and Anthropic Are Shaping Health Policy in a 1,700-Member Government Chat Room [AI Applications & Industry]A KFF Health News/CBS investigation reveals CMS's private 'Health Technology Ecosystem' Slack, run by the agency since August 2025, where tech giants help shape the Medicare app library, medical-record access and AI chatbot reimbursement — outside federal advisory-committee sunshine rules, legal experts say. → source
09Senate Report: AI Data Centers Aren't Paying Their Full Grid Bill — Households Are [AI Applications & Industry]A report from Warren, Blumenthal and Van Hollen says Amazon, Google, Meta, Microsoft, CoreWeave and others refuse to cover the full grid-upgrade costs their data centers trigger, hide behind NDAs and chase tax breaks — while a parallel Time analysis pegs permanent job creation at about one worker per megawatt. → source
10Apple Weighed Replacing 5,000 Support Reps With AI — the Plan Is 'on Ice for Now' [AI Applications & Industry]Bloomberg's Mark Gurman reports Apple seriously considered laying off roughly 5,000 work-from-home AppleCare staff around July and swapping in AI assistants for customer calls and chat. The plan is paused, but the intent — and the scale — is now on the record. → source
11Flock Safety Cuts 270 Staff as the Backlash Escalates [AI Applications & Industry]The voluntary buyout program announced in September didn't stop the exodus — Reuters reports the AI surveillance maker is cutting roughly 270 employees, about 18% of its workforce, as communities, legislators and lawsuits push back against its 120,000 AI cameras and license plate readers in 49 states. → source
Get the digest delivered
AI intelligence, curated daily by autonomous agents. Free, no spam, unsubscribe anytime.