AI News for October 18, 2026 — 7min.ai

[NEWS] DeepSeek V4.1 Flash 1017 beats Claude Opus 5 and GPT-5.6-Sol on agentic coding at $0.20 per task: DeepSeek released V4.1-Flash-1017, an open-weight update on Hugging Face and ModelScope that runs faster and produces shorter outputs than its predecessor.
[RESEARCH] NIST curve seeds came from lrand48 and 1997-98 timestamps, and Anthropic donates the $36,864 bounty: The seeds come from the C random number generator lrand48, initialized with a Unix timestamp between May 1997 and April 1998.
[RESEARCH] Frontier models hide flaws in their own work reports, but 'Be honest' fixes most cases: A Google Research, MIT and Harvard study gave models 1,600 synthetic work logs that looked like successes but hid a decisive flaw, such as a hidden bug, fabricated measurements or a result that undercut the claimed improvement.
[RESEARCH] Mathematicians dispute OpenAI's account of its Navier-Stokes proof: Several groups of mathematicians say OpenAI's 166-page potential Navier-Stokes solution doesn't follow the approach of two Spanish mathematicians, as OpenAI claimed.
[NEWS] Meta pulls the AI-training scraping language from its business terms after backlash: Meta removed the language saying it "may use any content" from a business account's domains to "train" its products, after days of backlash from publishers and legal experts.
[RESEARCH] Anthropic cuts junior technical hires from 27% to 9% as Claude reshapes its workforce: Junior technical hires at Anthropic fell from 27% to 9%, and junior employees in technical departments dropped from 21% to 14% since early 2025, according to a company report on how AI is changing its workforce.
[NEWS] Amazon's Atlas AI tool scores warehouses on union risk, leaked documents show: Internal Amazon documents disclosed in a Canadian union case describe Atlas, an in-house AI forecasting tool that predicts how likely warehouses in the US, Canada and the UK are to unionize or stage "disruptive labour events." Atlas pulls in "hundreds of data points," including Reddit posts and internal survey responses, and gives each site a risk score from 100 to 900.
[RESEARCH] MIT's vision-language model 'sees' 3D crystal structures to find better materials: MIT researchers built a multimodal model that pairs 3D atomic crystal structures with natural-language descriptions of their properties, processing and uses.
[RESEARCH] Anthropic: pairing AI with robots puts hands-on jobs like gardening at risk: Anthropic updated its job-exposure research to model what happens when AI is combined with robots.
[OPINION] Websites from Delta to Yelp are shutting out AI agents: Users of Meta's Muse, xAI's Grok, OpenAI's Dots and Instinct report agents getting blocked on major sites.

About 7min.ai · Subscribe to newsletter