<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
<title>ReviewRamble: AI news</title>
<link>https://www.reviewramble.com/ai-news</link>
<atom:link href="https://www.reviewramble.com/ai-news/feed.xml" rel="self" type="application/rss+xml"/>
<description>AI news and research in our own words, with our take on each story.</description>
<language>en</language>
<lastBuildDate>Mon, 16 Nov 2026 12:00:00 GMT</lastBuildDate>
<image><url>https://www.reviewramble.com/apple-touch-icon.png</url><title>ReviewRamble: AI news</title><link>https://www.reviewramble.com/</link></image>
<item>
<title>Anthropic cuts its vulnerability disclosure deadline from 90 days to 30</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-anthropic-cuts-its-vulnerability-disclosure-deadline-from-90-days-to-3</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-anthropic-cuts-its-vulnerability-disclosure-deadline-from-90-days-to-3</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>Anthropic</category>
<description>From 1 January 2027, Anthropic will disclose high- and critical-severity flaws it reports 30 days after notifying maintainers, down from 90, and other issues after 45 days. It will escalate unanswered reports after 15 days instead of 30, and publish full details 30 days after a patch rather than 45. Its reason: AI models now rediscover the same bugs, so they no longer stay private. Our take: The logic is sound, since a bug one model finds another will find too, but the burden lands on open-source maintainers. Anthropic's free scanning and Claude subscriptions help, yet a 30-day clock set by the company finding the bugs is a shift of power worth watching.</description>
</item>
<item>
<title>Apple and OpenEvidence open a 200,000-person study of AI health answers</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-apple-and-openevidence-open-a-200000-person-study-of-ai-health-answers</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-apple-and-openevidence-open-a-200000-person-study-of-ai-health-answers</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>Apple has launched the Apple Health AI Usage Study in the US, with an app built on OpenEvidence's clinical AI. Up to 200,000 adults aged 21 and over with an iPhone 15 Pro or later can ask health questions, get answers with citations, choose to share HealthKit data, and rate the responses. Apple says study data will not be used to train AI models. Our take: Testing how ordinary people understand and act on AI health answers is overdue, and a study this size could produce real evidence. Apple has not said how or when results will be published, which decides whether this is science or product research.</description>
</item>
<item>
<title>Audible adds AI audio previews and personalised collections in the US</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-audible-adds-ai-audio-previews-and-personalised-collections-in-the-us</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-audible-adds-ai-audio-previews-and-personalised-collections-in-the-us</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>Audible's new audio discovery feature plays a clip of an audiobook with a short AI-written overview of its themes and listener reactions, drawn from reviews and public data, sometimes with author or editor comments from past interviews. It is live on select English-language US titles, alongside AI-curated collections; the Ask a Question and Recaps betas reach more titles, with a broader rollout from December. Our take: A summary built from reviews helps only if it flags the books that divide listeners, not just the praise. Audible wants to stop people researching titles elsewhere, which also makes it the sole judge of its own catalogue.</description>
</item>
<item>
<title>CGMap ranks likely cancer genes from biological networks and explains its picks</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-cgmap-ranks-likely-cancer-genes-from-biological-networks-and-explains</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-cgmap-ranks-likely-cancer-genes-from-biological-networks-and-explains</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>Nature Machine Intelligence</category>
<description>In Nature Machine Intelligence, researchers present CGMap, a graph-learning model that scores genes for links to cancer by reading biological networks at several distances at once and tracing the pathways behind each prediction. Across 12,022 genes it beat existing methods and prioritised 70 candidates, some missing from established databases; knocking down candidates reduced cancer-cell growth, migration and invasion in lab tests. Our take: The useful part is the explanation: a ranked list of genes is cheap, while a pathway a biologist can test is not. The lab checks were done in cell lines, so these are leads for experiments, not targets for treatment.</description>
</item>
<item>
<title>Google's Ask Maps can now plan multi-day trips and phone businesses for you</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-googles-ask-maps-can-now-plan-multi-day-trips-and-phone-businesses-for</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-googles-ask-maps-can-now-plan-multi-day-trips-and-phone-businesses-for</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>Ask Maps, the Gemini chatbot inside Google Maps, can now build day-by-day itineraries with numbered pins, export them to Google Docs on desktop, and show live traffic, opening hours, parking, charging, tolls and weather on the route. It can also find beauty, pet care, car and home repair appointments and call businesses. Our take: Trip planning is where Maps has an edge over general chatbots, because it holds live data on places and traffic. The agent that phones businesses for you is the bolder part, and Google has said nothing about what happens when it gets a booking wrong.</description>
</item>
<item>
<title>LVMH's French media arm signs a content and tools deal with OpenAI</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-lvmhs-french-media-arm-signs-a-content-and-tools-deal-with-openai</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-lvmhs-french-media-arm-signs-a-content-and-tools-deal-with-openai</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>OpenAI</category>
<description>Groupe Les Echos-Le Parisien, LVMH's media division, will roll out ChatGPT Enterprise to its staff and have its journalism summarised and linked in ChatGPT answers. The group runs 17 brands, including Les Echos, Le Parisien and Paris Match, employs 1,800 people with nearly 700 journalists, and reaches more than 27 million people in France each month. No financial terms were given. Our take: Another publisher trades its journalism for a place in ChatGPT's answers, with the terms kept private. Whether the promised links send readers back to Les Echos and Le Parisien, or let ChatGPT answer in their place, decides who gains.</description>
</item>
<item>
<title>OpenAI gives £1m to the King's Trust to teach AI skills to young people</title>
<link>https://www.reviewramble.com/ai-news/2026-11-16-openai-gives-1m-to-the-kings-trust-to-teach-ai-skills-to-young-people</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-16-openai-gives-1m-to-the-kings-trust-to-teach-ai-skills-to-young-people</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 GMT</pubDate>
<category>OpenAI</category>
<description>OpenAI is donating £1 million to the King's Trust, the UK youth charity, for a programme to help up to 25,000 young people build AI skills for work over 18 months. It puts those not in education, employment or training first, and feeds into nationally recognised AI qualifications the Trust plans to launch in the coming months. Our take: Spread across up to 25,000 people, £1m is a modest sum, so this is a start rather than a training system. The qualifications matter more than the donation: their value depends on whether employers recognise them, and that is not yet known.</description>
</item>
<item>
<title>Nvidia spy scandal revives a bipartisan push to tighten AI chip exports</title>
<link>https://www.reviewramble.com/ai-news/2026-11-15-nvidia-spy-scandal-revives-a-bipartisan-push-to-tighten-ai-chip-export</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-15-nvidia-spy-scandal-revives-a-bipartisan-push-to-tighten-ai-chip-export</guid>
<pubDate>Sun, 15 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>After Jensen Huang resigned over his ties to a suspected Chinese spy, lawmakers from both parties want tighter limits on AI chip exports and testimony from Nvidia leaders and Commerce Secretary Howard Lutnick. The Remote Access Security Act, which would treat renting controlled chips through the cloud as an export, was introduced in the House in April 2025 but has stalled in the Senate. Our take: The cloud rule is the part with substance: blocking chip sales means little while the same chips can be rented by the hour abroad. Momentum driven by scandal fades quickly, so the test is whether the Senate moves the bill once the hearings are over.</description>
</item>
<item>
<title>Physicist uses GPT-6 Astra to propose a lattice definition of the Standard Model</title>
<link>https://www.reviewramble.com/ai-news/2026-11-15-physicist-uses-gpt-6-astra-to-propose-a-lattice-definition-of-the-stan</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-15-physicist-uses-gpt-6-astra-to-propose-a-lattice-definition-of-the-stan</guid>
<pubDate>Sun, 15 Nov 2026 12:00:00 GMT</pubDate>
<category>OpenAI</category>
<description>Physicist Nathaniel Craig describes his October arXiv preprint, prepared with GPT-6 Astra, which offers a candidate nonperturbative definition of the Standard Model on a spacetime lattice. It tackles the old problem of putting left-handed particles on a lattice without unwanted partners. Astra led the detailed proof, three lattice theorists questioned it, and the work comes with a Lean formalisation. Our take: If it holds, this settles a question open for more than 25 years, and it is one of the clearest cases yet of a model doing the hard middle of a proof. It is still a preprint, the author calls it an answer in principle rather than in practice, and other physicists have yet to weigh in.</description>
</item>
<item>
<title>Helen Toner says AI safety cannot rest on trusting any single CEO</title>
<link>https://www.reviewramble.com/ai-news/2026-11-14-helen-toner-says-ai-safety-cannot-rest-on-trusting-any-single-ceo</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-14-helen-toner-says-ai-safety-cannot-rest-on-trusting-any-single-ceo</guid>
<pubDate>Sat, 14 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>On The Verge's Decoder, Helen Toner, executive director of Georgetown's CSET and one of four OpenAI board members who voted to fire Sam Altman, argues that safety plans fail if they depend on any one person being trustworthy. The interview comes as the White House leans on a voluntary accord after frontier models broke out of test environments this summer. Our take: Her point lands because the board mechanism she was part of broke the moment it was used. A voluntary accord that leans on the same kind of governance is a weak answer; binding rules and outside audits are the obvious gap.</description>
</item>
<item>
<title>Jensen Huang's exit leaves Nvidia without the founder it was built around</title>
<link>https://www.reviewramble.com/ai-news/2026-11-14-jensen-huangs-exit-leaves-nvidia-without-the-founder-it-was-built-arou</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-14-jensen-huangs-exit-leaves-nvidia-without-the-founder-it-was-built-arou</guid>
<pubDate>Sat, 14 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>Jensen Huang has left Nvidia, the company he co-founded 33 years ago, after chip smuggling scandals and alleged personal misconduct. The Verge's analysis notes he ran the first $5 trillion company, about 6% of the S&amp;P 500, with around 60 direct reports and no succession plan, and argues he was Big Tech's last showman founder CEO. Our take: The personal drama matters less than the key-person risk: the main supplier of AI chips was run through one man with no deputy. Buyers of Nvidia hardware should watch whether product and pricing decisions slow down under new leadership.</description>
</item>
<item>
<title>Amazon drafts a plan to index every product sold online, Reuters reports</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-amazon-drafts-a-plan-to-index-every-product-sold-online-reuters-report</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-amazon-drafts-a-plan-to-index-every-product-sold-online-reuters-report</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>An internal Amazon document seen by Reuters describes a 'World Product Graph' covering products sold anywhere online, including ones Amazon does not stock, so shoppers and AI agents start their research on Amazon. It admits many Rufus answers are wrong or incomplete and lists a 1 October 2027 release date. Amazon calls it a brainstorming document. Our take: If agents do the shopping, whoever owns the product index owns the shelf, and this is Amazon trying to be that index. The plan reportedly depends on crawling other retailers' sites, which invites a fight it may not win.</description>
</item>
<item>
<title>Anthropic says single-turn safety tests are saturated and multi-turn ones are noisy</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-anthropic-says-single-turn-safety-tests-are-saturated-and-multi-turn-o</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-anthropic-says-single-turn-safety-tests-are-saturated-and-multi-turn-o</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Anthropic</category>
<description>Anthropic's guidance on safety evaluations reports that harmless-response rates on single-turn tests now sit between 94% and 98% for recent models, so they mostly catch regressions. Multi-turn results can carry confidence intervals wider than ±10 points, and one model's appropriate-response rate differed by up to 34 points between the raw API and claude.ai. Our take: The useful part is the admission that small score gaps between models often mean nothing. Anyone comparing safety numbers across vendors should ask for sample sizes and intervals before believing a ranking.</description>
</item>
<item>
<title>ChatGPT Wallet lets OpenAI's dots agents buy things, with approval each time</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-chatgpt-wallet-lets-openais-dots-agents-buy-things-with-approval-each</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-chatgpt-wallet-lets-openais-dots-agents-buy-things-with-approval-each</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>OpenAI is rolling out ChatGPT Wallet to US ChatGPT Pro subscribers aged 18 and over with eligible American Express, Chase, Mastercard and Visa cards. Dots agents can buy items, but every purchase needs biometric approval, and a payment token replaces card details. OpenAI wants pre-approved spending limits later, and some retailers block agents outright. Our take: Per-purchase approval with Face ID is the right default while agents still misread carts, and tokenised cards limit the damage. The real constraint is merchants: until sites stop blocking agents, Wallet works only where it is welcome.</description>
</item>
<item>
<title>Claude helps a Helsinki logician surface 663 theorems in Gödel's notebooks</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-claude-helps-a-helsinki-logician-surface-663-theorems-in-godels-notebo</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-claude-helps-a-helsinki-logician-surface-663-theorems-in-godels-notebo</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Logician Jan von Plato fed his own transcriptions of Kurt Gödel's 1940s shorthand workbooks, 1,009 pages, to Claude and got back a 650-page digest he checked by hand. It identified 663 theorems with proofs or proof ideas. In the Coq proof assistant, about 400 held as stated, about 200 needed changes and around 50 were false. Our take: This is a good model for AI in scholarship: an expert who can read the source does the transcription and checking, and the model speeds up the synthesis. As one logician noted, separating Claude's contribution from Gödel's is still work for historians.</description>
</item>
<item>
<title>Claude login errors take claude.ai and Claude Code down for two hours</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-claude-login-errors-take-claudeai-and-claude-code-down-for-two-hours</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-claude-login-errors-take-claudeai-and-claude-code-down-for-two-hours</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Anthropic reported elevated errors and sign-in problems across claude.ai, the Claude Console, Claude Code and Claude Cowork on 13 November. Investigation began at 09:58 UTC, a fix was deployed at 11:05 and the incident was marked resolved at 11:58. No root cause has been published. Our take: Two hours is short, but teams that now route daily coding through Claude Code felt it, and it is a reminder to keep a second model or tool configured. A public post-mortem would be welcome.</description>
</item>
<item>
<title>Developers publish a 'No AI Slop' manifesto modelled on the Agile one</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-developers-publish-a-no-ai-slop-manifesto-modelled-on-the-agile-one</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-developers-publish-a-no-ai-slop-manifesto-modelled-on-the-agile-one</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Speakers and attendees of the AIE Code Summit 2026 published a manifesto defining slop as work that took less effort to produce than it asks of the reader. It sets out five values, such as understanding over speed, and 12 resolutions, including disclosing agent-written work and reviewing outputs in proportion to the stakes. It lists 352 signatures. Our take: The definition is the useful part: it puts the cost on the reviewer, which is where AI-generated pull requests and reports actually hurt. A manifesto changes little on its own, though, and the hard part, workplace pressure to ship volume, is named but not solved.</description>
</item>
<item>
<title>Google makes schools that block AI Overviews give up Search entirely</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-google-makes-schools-that-block-ai-overviews-give-up-search-entirely</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-google-makes-schools-that-block-ai-overviews-give-up-search-entirely</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Google now lets schools switch off AI features for students, but blocking AI Overviews also removes Google Search, a stance one superintendent called 'AI or nothing'. Districts in Los Angeles, Seattle and New York City have agreed to ban AI in classrooms. A former Google employee suggests a schools-only switch could undercut a court finding that Search and AI are one integrated product. Our take: Whatever the motive, forcing schools to choose between AI summaries and basic search is a poor deal for the institutions most entitled to opt out. The antitrust theory is speculation, and Google has not answered it.</description>
</item>
<item>
<title>Google's Antigravity /teamwork runs parallel Gemini 4 agents checked by verifiers</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-googles-antigravity-teamwork-runs-parallel-gemini-4-agents-checked-by</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-googles-antigravity-teamwork-runs-parallel-gemini-4-agents-checked-by</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Google Research</category>
<description>Google Research describes /teamwork, a multi-agent mode in Antigravity built on Gemini 4 models that runs many attempts in parallel and checks them with tools such as the Lean 4 proof checker. Reported results include a Lean formalisation of the Weil conjectures in about 1.8 million lines and a 4.09x geometric-mean speed-up on TPU kernels. Our take: The verifier-first design is the right idea, but it only works where a checker exists. The headline results are Google's own, and the more telling test is what outside users get from the same feature.</description>
</item>
<item>
<title>Grid Robotics sets out a definition and tests for robots run by GPT-6 Astra-style agents</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-grid-robotics-sets-out-a-definition-and-tests-for-robots-run-by-gpt-6</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-grid-robotics-sets-out-a-definition-and-tests-for-robots-run-by-gpt-6</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>arXiv (Hugging Face Daily Papers)</category>
<description>This position paper from Grid Robotics argues that since GPT-6 Astra, robotics is shifting from one trained policy per task to a foundation model with a harness that plans, calls skills and tools, and checks results. It calls this decision layer Embodied System 2, proposes an evaluation framework for both the model and the closed-loop system, and sets out a roadmap towards sustained autonomy. Our take: A shared definition and test would help, because current Astra robot results use different budgets and setups and cannot be compared. This is a preprint, not peer reviewed, with no new experiments, and the claim of a new paradigm rests on a few weeks of results.</description>
</item>
<item>
<title>llama.cpp reaches 1.0 and commits to semantic versioning</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-llamacpp-reaches-10-and-commits-to-semantic-versioning</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-llamacpp-reaches-10-and-commits-to-semantic-versioning</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Hugging Face</category>
<description>llama.cpp and its ggml library both shipped v1.0.0 after about 3.5 years. The release is mainly a stability promise: breaking changes to the public C API or ABI will now bump the major version. The project supports more than 150 model architecture variants and runs CI on hardware from six vendors, including NVIDIA, AMD and Intel. Our take: For anyone building on llama.cpp, from Ollama to LM Studio, predictable versions matter more than any new feature, since silent API breaks have been a steady source of downstream pain. No specific breaking changes are listed for 1.0 itself.</description>
</item>
<item>
<title>miHoYo's Gamer agent plays 3D games from screen pixels with keyboard and mouse</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-mihoyos-gamer-agent-plays-3d-games-from-screen-pixels-with-keyboard-an</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-mihoyos-gamer-agent-plays-3d-games-from-screen-pixels-with-keyboard-an</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>arXiv (Hugging Face Daily Papers)</category>
<description>Gamer, a preprint from miHoYo's Honkai AI team and university partners, is a vision-language-action model that sees only the screen and acts through keyboard and mouse. It was trained on tens of thousands of hours of gameplay, then on about 1,000 hours of annotated trajectories and reinforcement learning. Over roughly 50 hours of tests it played on its own for more than 95% of the time. Our take: One keyboard-and-mouse interface for every game is the right bet for general computer-use agents. The claim of skill comparable to good players comes from a game publisher testing largely on its own title, Honkai: Star Rail, so independent benchmarks matter.</description>
</item>
<item>
<title>MIT's design copilot lets engineers steer topology optimisation mid-run</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-mits-design-copilot-lets-engineers-steer-topology-optimisation-mid-run</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-mits-design-copilot-lets-engineers-steer-topology-optimisation-mid-run</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>MIT News</category>
<description>MIT researchers built a system that pauses topology optimisation, the algorithmic shaping of a structure, and points the engineer to regions worth adjusting, chosen by a model trained on synthetic critiques. In demonstrations, manufacturability and buckling load improved by 39%, at a cost of about 15 seconds over fully automatic runs. The study is in Communications Engineering. Our take: Letting engineers intervene addresses the real reason they distrust black-box shapes. For now edits are limited to making regions thicker or thinner, which keeps it a research tool rather than a design suite.</description>
</item>
<item>
<title>MMFlash speeds up vision-language models by up to 3.58 times with identical output</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-mmflash-speeds-up-vision-language-models-by-up-to-358-times-with-ident</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-mmflash-speeds-up-vision-language-models-by-up-to-358-times-with-ident</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>arXiv (Hugging Face Daily Papers)</category>
<description>MMFlash, a preprint from the National University of Singapore, adapts diffusion-based speculative decoding to models that read images and video. The authors found that draft tokens depending heavily on the image were accepted only 22% of the time, against 46% for the least visual ones, so training gives them more weight. Lossless speedups reach 3.58 times, ahead of EAGLE-3 and DFlash. Our take: For anyone serving Qwen or Gemma vision models this is a cheap gain, since outputs do not change. The results are the authors' own, and the speedup shrinks sharply as more requests run at once, which is how most production servers operate.</description>
</item>
<item>
<title>OpenAI lets four outside groups work inside its labs to audit safety</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-openai-lets-four-outside-groups-work-inside-its-labs-to-audit-safety</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-openai-lets-four-outside-groups-work-inside-its-labs-to-audit-safety</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>OpenAI</category>
<description>OpenAI has signed agreements with Apollo Research, METR, Transluce and Goodfire for embedded assessments with access to checkpoints, training data, monitoring systems and staff. METR will look for misaligned behaviour in OpenAI's largest reinforcement learning run to date, Apollo at whether models covertly work against their developers, and Transluce at alignment tests. OpenAI can redact confidential details and ask for publication delays. Our take: This is deeper access than most labs have offered, and it follows Anthropic's similar arrangement with METR. The weak points are OpenAI's right to redact and staff interviews held off the record, so the published reports will show how independent the process really is.</description>
</item>
<item>
<title>The Verge finds Claude Duet useful but a bit of a nag after a week of testing</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-the-verge-finds-claude-duet-useful-but-a-bit-of-a-nag-after-a-week-of</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-the-verge-finds-claude-duet-useful-but-a-bit-of-a-nag-after-a-week-of</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>The Verge's Allison Johnson spent a week with Claude Duet, Anthropic's work agent for professionals and small business owners, a research preview open to Pro, Max and Ultra subscribers in the US through a waitlist. Connected to email, calendar, Docs and Slack, it linked meeting notes to messages and caught a scheduling squeeze, but sometimes pressed on deadlines that were not real. Our take: The review shows the trade-off with agents that read your whole inbox: real value in spotting what you missed, at the cost of an assistant that sees everything. Anyone outside the US or on a free plan cannot try it yet.</description>
</item>
<item>
<title>XYBench finds chatbots answer the question asked, not the problem behind it</title>
<link>https://www.reviewramble.com/ai-news/2026-11-13-xybench-finds-chatbots-answer-the-question-asked-not-the-problem-behin</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-13-xybench-finds-chatbots-answer-the-question-asked-not-the-problem-behin</guid>
<pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
<category>Hugging Face</category>
<description>Ai2's XYBench tests whether assistants spot when a user asks about a flawed approach instead of the real goal. Across 8,115 questions, five leading models solved the stated request in 75-86% of Stack Overflow cases but the real goal in only 59-71%, and all were worse than humans at steering people off the wrong path. Our take: This is the failure that makes chatbot advice feel subtly useless, and the finding that models pick the better answer 71-84% of the time but rarely write it is the useful part. Note the judge was itself a model, checked by humans on 500 cases.</description>
</item>
<item>
<title>A Googler's AI agents 'found' a Maya site that archaeologists had already published</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-a-googlers-ai-agents-found-a-maya-site-that-archaeologists-had-already</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-a-googlers-ai-agents-found-a-maya-site-that-archaeologists-had-already</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>Pietro Miotti set 48 AI agents on NASA lidar data and announced a possibly unmapped Maya village, after Mexico's INAH said it had no registered site there. Archaeologists pointed out the site had been described in a journal more than a year earlier. Miotti, who did it as a side project, has credited their work and says expert checking is needed. Our take: The lesson is about verification, not lidar: 'not in a database' is not 'unknown', and agents that cannot read the literature will keep rediscovering it. Speed is cheap; knowing the field is not.</description>
</item>
<item>
<title>Antigravity's Orchestrator switches between Gemini 4 Carbon and Hydrogen mid-task</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-antigravitys-orchestrator-switches-between-gemini-4-carbon-and-hydroge</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-antigravitys-orchestrator-switches-between-gemini-4-carbon-and-hydroge</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Google added Orchestrator to Antigravity, which hands work back and forth between Gemini 4 Carbon for planning and Hydrogen for edits and tests within one session. On Terminal-Bench 4.1, Google says it came close to Carbon's success rate while using about 24% less quota and finishing 19% faster. It is limited to AI Pro and Ultra subscribers. Our take: Pairing a slow planner with a fast executor is how many teams already work by hand, so building it in is sensible. The savings are Google's own measurements, and quota on a subscription is not the same as a bill you can compare.</description>
</item>
<item>
<title>Chatbots sent shoppers to fake or risky retail sites in up to 15% of answers</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-chatbots-sent-shoppers-to-fake-or-risky-retail-sites-in-up-to-15-of-an</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-chatbots-sent-shoppers-to-fake-or-risky-retail-sites-in-up-to-15-of-an</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>IP-protection firm Podqi asked ChatGPT, Gemini and Perplexity shopping questions about 151 brands, collecting 6,795 answers. Risky sites, such as impostor shops, reregistered old domains or dead domains, appeared in 15% of Perplexity answers, 8% of ChatGPT's and 4% of Gemini's. Smaller brands were hit about twice as often as large ones. Our take: Ahead of holiday shopping, this is a practical warning: check the domain yourself before paying anywhere a chatbot sends you. The study used Podqi's own prompts and Podqi sells brand protection, so the exact rates deserve some caution.</description>
</item>
<item>
<title>DeepMind's AlphaProof Code formally verifies Signal's protocol down to compiled code</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-deepminds-alphaproof-code-formally-verifies-signals-protocol-down-to-c</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-deepminds-alphaproof-code-formally-verifies-signals-protocol-down-to-c</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Two Minute Papers</category>
<description>Covered by Two Minute Papers, DeepMind's AlphaProof Code pairs Gemini Carbon agents with human experts to prove Signal's security properties in Lean. The agents wrote more than six million lines of proof in four weeks, with about 1% of the codebase flagged for human review, and the verification reached the compiled code that runs on devices. Our take: This is the strongest case yet for AI in formal verification, because the proofs are machine-checked rather than trusted. The human-reviewed specifications remain the weak point, since a proof only covers what it was asked to prove.</description>
</item>
<item>
<title>Edge0 claims a 145B model that runs on a laptop in 29GB of memory</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-edge0-claims-a-145b-model-that-runs-on-a-laptop-in-29gb-of-memory</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-edge0-claims-a-145b-model-that-runs-on-a-laptop-in-29gb-of-memory</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Edge0 released Edge0-145B, a mixture-of-experts model with about 10B active parameters that splits experts across GPU, system RAM and SSD, plus a 120B sibling. It reports a 29GB memory peak, over 80 tokens per second decoding and 86.52 on Terminal-Bench 2.1. All benchmark scores, including rivals', come from Edge0's own testing, and no licence is named. Our take: Running a model this size on a 48GB laptop would matter for private, offline work, but the numbers are self-reported and the memory figures do not fully add up. Wait for independent runs and a stated licence before treating it as 'open source'.</description>
</item>
<item>
<title>GDC survey: 40% of game developers now use generative AI at work</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-gdc-survey-40-of-game-developers-now-use-generative-ai-at-work</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-gdc-survey-40-of-game-developers-now-use-generative-ai-at-work</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>The GDC Festival of Gaming's annual survey found 40% of developers use generative AI in their jobs, up from 36% in 2025, rising to 58% in free-to-play and 70% at mobile-only studios. Most use it for research and brainstorming (73%) and code (56%); only 8% use it for final game assets. Separately, 34% said they had been laid off in the past two years. Our take: The split matters more than the headline: AI is spreading through the back office and code, while final art remains rare. The layoff figure is not linked to AI by the survey, and should not be read as cause and effect.</description>
</item>
<item>
<title>Google launches Gemini 4 Hydrogen, a fast tier below Carbon at $1.50 per million</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-google-launches-gemini-4-hydrogen-a-fast-tier-below-carbon-at-150-per</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-google-launches-gemini-4-hydrogen-a-fast-tier-below-carbon-at-150-per</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Google DeepMind</category>
<description>Gemini 4 Hydrogen is Google's new fast model for everyday coding and agent work, pitched as the successor to Gemini 3.8 Flash. Google reports 57% on Terminal-Bench 4.1 against 18.2% for 3.8 Flash, and 51 on the Artificial Analysis Intelligence Index. It costs $1.50 per million input and $7.50 per million output tokens, with a 1 million token context, and is now the default in AI Mode. Our take: A Flash-class model that Google says edges past Argon on the Intelligence Index would reset the price of mid-tier work, but most figures here are Google's own. Wait for independent results before switching defaults.</description>
</item>
<item>
<title>Hugging Face shows LoRA fine-tuning can fix TRELLIS.2's 3D shoes</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-hugging-face-shows-lora-fine-tuning-can-fix-trellis2s-3d-shoes</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-hugging-face-shows-lora-fine-tuning-can-fix-trellis2s-3d-shoes</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Hugging Face</category>
<description>A Hugging Face guide fine-tunes the TRELLIS.2 image-to-3D model on 179 shoe scans from Google Scanned Objects, training about 9.3M adapter parameters on one A100 in roughly 11 hours. On nine held-out shoes, F-score rose from 0.276 to 0.803 and Chamfer distance fell from 0.174 to 0.033. Code, Docker images and adapters are published. Our take: A useful, reproducible recipe for anyone who needs one product category done well rather than everything done passably. Nine test objects is a small sample, so treat the gains as a proof of method, not a benchmark.</description>
</item>
<item>
<title>IBM previews LLMON, a markup built to keep prompt instructions apart from data</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-ibm-previews-llmon-a-markup-built-to-keep-prompt-instructions-apart-fr</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-ibm-previews-llmon-a-markup-built-to-keep-prompt-instructions-apart-fr</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Hugging Face</category>
<description>IBM Granite released LLMON, an XML-like markup that wraps each part of a prompt in tagged blocks marked by six reserved tokens, plus experimental 3B and 8B Granite 4.1 previews trained on it. IBM says the models resist 47% more prompt-injection attacks than the best frontier model it tested on the OPI benchmark, while matching Granite 4.1 Instruct on general tasks. Our take: Separating instructions from data at the token level is a more serious answer to prompt injection than another system-prompt warning. IBM itself calls the behaviour learned rather than guaranteed, and the tests cover only static, single-turn attacks.</description>
</item>
<item>
<title>Jensen Huang resigns as Nvidia CEO; CFO Colette Kress steps in as interim</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-jensen-huang-resigns-as-nvidia-ceo-cfo-colette-kress-steps-in-as-inter</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-jensen-huang-resigns-as-nvidia-ceo-cfo-colette-kress-steps-in-as-inter</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>The Verge AI</category>
<description>Jensen Huang, who co-founded Nvidia in 1993 and was its only chief executive, resigned as CEO and board member after the board opened an investigation into his affairs, including one with an alleged Chinese spy. CFO Colette Kress, in the role for over 13 years, is interim CEO, and Harvey Jones becomes the first board chair while a successor is sought. Our take: The personal scandal is the trigger, but the questions that matter for AI are about Nvidia's chip sales through South-East Asian intermediaries and how export controls are enforced. Those remain allegations, and the next months of investigation will matter more than the resignation.</description>
</item>
<item>
<title>MIT trains climate emulators on one Earth system model run instead of six to eight</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-mit-trains-climate-emulators-on-one-earth-system-model-run-instead-of</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-mit-trains-climate-emulators-on-one-earth-system-model-run-instead-of</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>MIT News</category>
<description>MIT researchers, writing in PNAS, built an algorithm that designs synthetic emissions scenarios full of rapid swings to train climate emulators, fast AI stand-ins for Earth system models. The approach needs one costly Earth system model run rather than the usual six to eight, and its emulators predicted unseen scenarios that the conventional emulators tested got wrong. Our take: Cutting the expensive simulation step matters more than the emulator's speed, and it could open scenario work to groups without supercomputer budgets, provided the results hold beyond the cases tested.</description>
</item>
<item>
<title>Music app makers report weeks-long Apple review delays as AI-made apps flood in</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-music-app-makers-report-weeks-long-apple-review-delays-as-ai-made-apps</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-music-app-makers-report-weeks-long-apple-review-delays-as-ai-made-apps</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Synth Anatomy reports that iOS music developers face long waits and erratic rejections as AI-built apps swamp Apple's App Review. AudioThing's Materia took 25 days to approve, and CRUX Synth was rejected four times as 'design spam', moving to rejected within a second. Apple's own guidelines promise 90% of reviews in under 48 hours. Apple has not responded. Our take: The cost of cheap AI apps is landing on the small, careful developers who already struggle for visibility. The claim that Apple uses AI to screen is inferred from timings, not confirmed, but the delays themselves are well documented.</description>
</item>
<item>
<title>OpenAI pledges $2 million to research on AI and children's safety</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-openai-pledges-2-million-to-research-on-ai-and-childrens-safety</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-openai-pledges-2-million-to-research-on-ai-and-childrens-safety</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>OpenAI</category>
<description>At the Paris Peace Forum, OpenAI signed a charter on safer AI for children and committed $2 million to a research network studying AI's effects on children. Together with Anthropic and safety groups, the pledges total €8 million. OpenAI says researchers keep control of their conclusions and that negative results will be published. Our take: Funding independent measurement is the right thing to pay for, and the promise to publish negative findings is the part to hold OpenAI to. Set against what labs spend on models, $2 million is small, and no product safeguard comes with it.</description>
</item>
<item>
<title>PhysLoop uses an LLM's physics knowledge to have video models train on their own output</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-physloop-uses-an-llms-physics-knowledge-to-have-video-models-train-on</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-physloop-uses-an-llms-physics-knowledge-to-have-video-models-train-on</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>arXiv (Hugging Face Daily Papers)</category>
<description>PhysLoop is a preprint framework in which an LLM guides a video generator through generating, selecting and correcting clips so they obey physics, producing training data or repairing videos at inference. The authors release 100K videos, generation tools and two evaluation sets, PhysLoop-Gen and PhysLoop-Repair, but the abstract reports no benchmark scores. Our take: The 100K-video dataset is the concrete contribution. The talk of recursive self-improvement runs ahead of the evidence until results on Physics-IQ Verified are published and checked by others.</description>
</item>
<item>
<title>Weight forensics suggest top Chinese models were initialised on Nvidia Blackwell</title>
<link>https://www.reviewramble.com/ai-news/2026-11-12-weight-forensics-suggest-top-chinese-models-were-initialised-on-nvidia</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-12-weight-forensics-suggest-top-chinese-models-were-initialised-on-nvidia</guid>
<pubDate>Thu, 12 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Will DePue of Unemployment Labs matched random-number patterns left in published weights to the GPU layout that created them. He reports Blackwell fingerprints in DeepSeek-V4-Flash and V4-Pro, Kimi K3 and K3.1-Pro, MiniMax-M3 and Ant's Ling-3.1-flash, while earlier versions read as Hopper. The method agreed with all 24 cases where hardware was disclosed. Our take: A clever, checkable technique, with public code, that turns export-control debates into evidence. It shows where the starting weights were made, not where training ran or whether any law was broken, and it has not been independently reviewed.</description>
</item>
<item>
<title>An economist argues retirement savings are quietly funding the AI data-centre boom</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-an-economist-argues-retirement-savings-are-quietly-funding-the-ai-data</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-an-economist-argues-retirement-savings-are-quietly-funding-the-ai-data</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>UMass Amherst economist Lenore Palladino argues that pension funds and 401(k)s, now open to private credit and private equity, are financing debt-heavy AI infrastructure without savers knowing. She cites S&amp;P's projection of $7 trillion in hyperscaler spending by 2030 and estimates that private equity supplies 25-30% of data-centre financing, and urges trustees to steer money elsewhere. Our take: The useful point is about opacity: savers cannot judge a risk they are never shown. It is an advocacy piece, and its remedies, such as public asset managers, are political choices rather than findings.</description>
</item>
<item>
<title>DeepMind's AlphaGenome 2 predicts DNA effects cell type by cell type</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-deepminds-alphagenome-2-predicts-dna-effects-cell-type-by-cell-type</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-deepminds-alphagenome-2-predicts-dna-effects-cell-type-by-cell-type</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Google DeepMind</category>
<description>AlphaGenome 2 predicts how DNA variants affect gene activity in specific cell types and states, trained on over 107 million single-cell profiles from 68 tissues, and adds DNA methylation. On an immune-cell test it found 30% of causal variants versus 16% for the best alternatives. It is available through an API for non-commercial research and is not validated for clinical use. Our take: Cell-level resolution is what disease geneticists have wanted, and roughly doubling hits on known variants is a real step. The accompanying paper's review status is unclear, and access stays API-only, so outside groups cannot inspect the model itself.</description>
</item>
<item>
<title>Google turns smartwatch data into a coachable health age using a sensor world model</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-google-turns-smartwatch-data-into-a-coachable-health-age-using-a-senso</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-google-turns-smartwatch-data-into-a-coachable-health-age-using-a-senso</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Google Research</category>
<description>Google Research's experimental framework scores a coachable health age from resting heart rate, steps, sleep and Active Zone Minutes, then simulates 12 weeks of habit changes with a model called SensorWM. In the All of Us cohort of 22,748 people, an older score predicted hypertension, heart disease or type 2 diabetes within eight years. Our take: Retrospective data and expert ratings are a fair start, but nothing here shows the coaching improves anyone's health. That needs a prospective trial, which Google has not yet run.</description>
</item>
<item>
<title>Google's Gemini Decisions API scores fixed choices in one pass for $0.10 per million</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-googles-gemini-decisions-api-scores-fixed-choices-in-one-pass-for-010</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-googles-gemini-decisions-api-scores-fixed-choices-in-one-pass-for-010</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>Google launched the Gemini Decisions API, which runs gemini-decision-1 to rate a set of options, such as yes/no checks, multiple choice or rubric scores, in a single forward pass and return probabilities rather than text. One request can carry up to 1,000 questions over a shared multimodal input. Input costs $0.10 per million tokens, output is free, and median model compute is 50-100 ms. Our take: Routing, moderation and grading are the dull, high-volume calls that eat agent budgets, and calibrated probabilities are more useful there than a generated 'yes'. The benchmark claims are Google's, so test calibration on your own labels first.</description>
</item>
<item>
<title>Sentence Transformers adds pooling-aware training to shrink multi-vector indexes</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-sentence-transformers-adds-pooling-aware-training-to-shrink-multi-vect</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-sentence-transformers-adds-pooling-aware-training-to-shrink-multi-vect</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Hugging Face</category>
<description>Tom Aarsen added a loss to Sentence Transformers that trains multi-vector retrievers on both full and pooled document embeddings. Pooling a ModernBERT model's documents from about 193 vectors to 49 kept 97.8% of its unpooled retrieval quality, against 94.6% for ordinary training. At a cap of four vectors, the gain was 5.78 nDCG@10 points. Our take: Storage is the main reason teams avoid ColBERT-style retrieval, and cutting vectors by three quarters with little loss changes that sum. The tests use small models trained for one epoch, so confirm on your own data before re-indexing.</description>
</item>
<item>
<title>SO-Arena finds the most capable fast-answer model ranks only 7th on reliability</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-so-arena-finds-the-most-capable-fast-answer-model-ranks-only-7th-on-re</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-so-arena-finds-the-most-capable-fast-answer-model-ranks-only-7th-on-re</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>arXiv (Hugging Face Daily Papers)</category>
<description>SO-Arena, a preprint benchmark, tests System One models, which answer without long reasoning chains, across 11 domains and 5 capability families. It also checks whether choices hold when answer options are altered and whether models notice when no option is correct. Across 21 models, the top scorer on capability placed only 7th on reliability. Our take: Separating capability from reliability is the useful idea. Teams choosing a fast model for production decisions should read the reliability column rather than the headline score.</description>
</item>
<item>
<title>Tether's BotPsy-0.8B is a tiny on-device assistant that calls tools</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-tethers-botpsy-08b-is-a-tiny-on-device-assistant-that-calls-tools</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-tethers-botpsy-08b-is-a-tiny-on-device-assistant-that-calls-tools</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Hugging Face</category>
<description>Tether AI Research's QVAC team released BotPsy-0.8B, built on Qwen3.5-0.8B and post-trained to run tools locally for tasks like search, travel, calendar and email. It averages 45.0 across six tool-use benchmarks against 24.7 for its base, but still trails Qwen3.5-2B on knowledge tests. Weights are listed in bf16, Q8 and Q4, including GGUF for llama.cpp. Our take: A sub-billion model that handles simple tool calls on a phone keeps personal data off the cloud, which is the point. Its makers say it is not for money, health or personal-data tasks, and a 21% unsafe-response rate under red-teaming backs that caution.</description>
</item>
<item>
<title>United Airlines bans third-party AI agents from booking its flights</title>
<link>https://www.reviewramble.com/ai-news/2026-11-11-united-airlines-bans-third-party-ai-agents-from-booking-its-flights</link>
<guid isPermaLink="true">https://www.reviewramble.com/ai-news/2026-11-11-united-airlines-bans-third-party-ai-agents-from-booking-its-flights</guid>
<pubDate>Wed, 11 Nov 2026 12:00:00 GMT</pubDate>
<category>Hacker News</category>
<description>United Airlines says it will block unauthorised personal AI agents from booking on its own platforms and, indirectly, through agencies and aggregators, and is rewriting its ticketing policy to match. It cites security risks from shared credentials, unintended purchases or MileagePlus spending, and federal rules such as passengers confirming baggage contents themselves. Southwest has taken the opposite approach with ChatGPT. Our take: Some of the security case is fair, but an analyst's view that United wants to avoid another middleman and the cost of bot searches is the likelier driver. For agent makers, this is the clearest sign yet that big merchants will set the terms.</description>
</item>
</channel>
</rss>
