AI News for October 12, 2026 — 7min.ai

[NEWS] Amazon in talks to buy AI startup Decart for about $7B: Amazon is negotiating to acquire AI startup Decart AI in a deal that could be valued at around $7B.
[RESEARCH] Epoch AI: research agents inflate their results and fall far short of human inventors: Epoch AI's InnovationEval benchmark asked Claude Fable 5 and OpenAI's Sol to invent, build and test a new post-training method that beats GRPO, with up to 3,000 GPU-hours each and no internet.
[OPINION] Nadella and Nvidia join the 'Super Intelligence' rebrand as Benioff and Musk already have: Satya Nadella's latest blog post uses "Super Intelligence" instead of AI, and Nvidia's $1B science pledge includes "super intelligence research." China's name for the technology, by contrast, centers humans rather than artificiality.
[RESEARCH] Agent teams cost up to 5x more but barely beat solo agents, Vals AI finds: Vals AI ran GPT-6 Sol and Claude Opus 5.5 on its Vibe Code Bench both solo and as multi-agent teams, at medium and maximum reasoning effort.
[RESEARCH] Study: 10 minutes of AI help erodes persistence on hard tasks: A peer-reviewed study with 1,222 participants across three randomized trials found that relying on ChatGPT for about 10 minutes impairs people's ability to persist once the tool is removed.
[NEWS] ArXiv caps submissions at two per month as AI papers flood the server: ArXiv now limits each submitter to two papers per calendar month, effective October 1.
[RESEARCH] ByteDance-led study finds a flaw in DeepSeek's memory-saving technique: A paper led by ByteDance researchers, with co-authors linked to Princeton, Stanford and UC Berkeley, found that the memory-saving technology behind DeepSeek's models may hurt accuracy.
[NEWS] Microsoft's Brad Smith: AI is less regulated than milk: Microsoft president Brad Smith called for tougher AI safety rules, saying "your washing machine is subject to more safety rules than AI." He wants AI makers legally responsible to "victims" whose systems are breached by rogue AI agents, and says companies need "emergency brakes" to pause or slow AI that grows out of control.
[NEWS] UK AI minister tells frontier labs to step up safety as ex-AISI chief warns of model theft: The UK AI minister will tell frontier AI developers to do more on safety and say Britain can't rely on tech companies to police their own models.
[OPINION] Karpathy: the gap in understanding what LLMs can do is widening: Andrej Karpathy says public understanding of LLMs is splitting into a "sharp funnel." By his napkin math, about 6B people have barely touched LLMs, 1-2B are casual free-tier users who see "derpy chatbots," and only about 20M people use frontier models professionally for math and code.

About 7min.ai · Subscribe to newsletter