Skip to content

Search stories, models, companies…Subscribe

The river

Latest

Every story as it’s published, newest first.

Thursday, Oct 29

30 stories

Models · NewsGemini 4 Carbon: $2/$10 at launch, then $4/$20, but only cyber defenders get it for nowGemini 4 Carbon, announced by Google on October 28, 2026, will launch at Argon's $2/$10 introductory price per million tokens, then $4/$20. For now it goes only to a subset of cyber defenders through Fairwind, with no public release date, model ID or model card.14:02·6 min read

Models · ExplainerGemini 4 Carbon vs Claude Opus 5.5 vs GPT-6.1 Sol: price, context, access and benchmarks (October 2026)Gemini 4 Carbon vs Opus 5.5 vs GPT-6.1 Sol, on official prices, context limits and access as of October 29, 2026, plus the shared benchmarks: Artificial Analysis scores them 59, 58 and 55. Carbon isn't on sale yet.14:02·9 min read

Industry · NewsAlphabet Q3 2026 earnings: $51.4 billion of capex, Google Cloud up 103%, a 1 billion-user Gemini app and $195–205 billion guidance unchangedAlphabet's Q3 2026 earnings, reported October 28, 2026: revenue of $127.6 billion, Google Cloud revenue up 103% to $30.7 billion, $51.4 billion of capital spending in the quarter and an unchanged 2026 capex range of $195–205 billion. Pichai said the Gemini app reached 1 billion users and Cloud's backlog was $561 billion.14:02·6 min read

Industry · NewsMicrosoft earnings: AI capex of $45.2 billion, Azure up 48%, and a 2026 spending view trimmed to about $170 billionMicrosoft's earnings for the September quarter, reported October 28, 2026: revenue of $92.3 billion, Azure growth of 48%, AI capex of $45.2 billion including finance leases, and a calendar 2026 spending view of about $170 billion after third-party capacity delays. What Microsoft said about OpenAI and its buyout offer.14:02·6 min read

Research · BenchmarksArtificial Analysis Intelligence Index v5: 60% private tests, three evaluations dropped, and Gemini 4 Carbon on top at 59Artificial Analysis Intelligence Index v5, launched October 27, 2026, rebuilds the index around nine evaluations, 60% of them with private test sets. Claude Opus 5.5 led at 58 until Gemini 4 Carbon scored 59 the next day. What changed, who leads, and why older scores don't compare.14:02·8 min read

Policy · NewsReligious exemption from AI: a University of Kentucky employee, a humanist, was excused from using AI at work. What's known, and what Title VII saysA religious exemption from AI at work: the American Humanist Association says the University of Kentucky agreed on September 25, 2026 that research administrator David Durman, a humanist, need not personally use AI. It was an internal accommodation, not a court ruling. What's known, and how Title VII and Groff v. DeJoy apply.14:02·6 min read

Research · BenchmarksAI agents tax benchmark: Surge AI's Dayjob: Tax finds the top agent passes 10.8% of attempts on 50 professional tax assignmentsSurge AI's Dayjob: Tax, an AI agents tax benchmark published October 27, 2026, gave 16 models 50 expert-built tax assignments. Claude Sonnet 5.5 led with a 10.8% pass rate; four models passed 0.4% or less. An LLM judge graded the work against expert rubrics, and only 2 of the 50 tasks are public.14:02·9 min read

Hardware · ExplainerMac Studio M5 Ultra 512GB: reported at $22,299, with 1.2TB/s of bandwidth, and which of the biggest open models actually fitThe Mac Studio M5 Ultra with 512GB of unified memory went on sale on October 28, 2026, at $22,299 according to MacRumors and Cult of Mac. What 512GB and 1.2TB/s mean for running the biggest open models locally: Reflection Beam fits at 4-bit, DeepSeek V4.1 Flash barely does, and Kimi K3.1 needs several machines.14:02·8 min read

Developers · ExplainerCursor Agent SDK explained: agents as a folder of code, no separate price, and how it differs from the Claude Agent SDK and OpenAI's Agents SDKCursor's Agent SDK, generally available since October 20, 2026, lets teams define an agent as a folder of Markdown and TypeScript, test it with evals and deploy it to Cursor or their own servers. Cursor publishes no separate price for it. Here's what it is, what it costs, and how it compares with the Claude Agent SDK and OpenAI's Agents SDK.14:02·8 min read

Products · AnalysisSteam's AI-generated games policy, and what Valve, Microsoft and Epic actually said about AI clones, 'AI slop' and the 'AI is theft' fightSteam's AI-generated games policy asks developers to disclose AI content that ships in a game or is generated while it runs; tools used only for efficiency don't count. One count puts a third of 2026 Steam releases under that label. What Valve told developers worried about AI clones on October 27, 2026, and why Kotaku argues Microsoft and Epic now treat AI copies of their games as theft.14:02·8 min read

Developers · AnalysisStack Overflow questions per month: the 'fewer than 1,000' claim, what the data query counts, and what Stack Overflow says about AIA Hacker News post on October 28, 2026 said Stack Overflow now gets fewer than 1,000 questions a month, citing a Stack Exchange Data Explorer query. The official Stack Exchange API counts 969 in September 2026. Here's what that kind of count includes and leaves out, why the latest month can look lower than it is, what earlier reports said, and what Stack Overflow's own survey says about AI.14:02·9 min read

Products · NewsGoogle AI and the 2020 election: AI Mode in the US pointed users to official records instead of naming Biden, and Google says it's 'working to address it'Google AI and the 2020 election: on October 28, 2026, Fast Company and TheWrap reported that Google Search's AI Mode in the US wouldn't name the winner and pointed to the FEC and National Archives, while the Gemini app answered. Google says it's working to address it. Biden won, with 306 electoral votes.14:02·6 min read

Open source · NewsNemotron 3.5 Transcribe: Nvidia's open speech recognition model covers 40 languages under OpenMDW, with about 4B active parametersNemotron 3.5 Transcribe, released by Nvidia on Hugging Face on October 28, 2026, is an open-weight speech recognition model for 40 languages under the OpenMDW-1.1 license. It activates about 4B parameters per token, runs on a single Nvidia GPU and handles recorded audio, not live streams. Nvidia's word error rates are vendor-reported.14:02·6 min read

Research · AnalysisQwenputer: a developer compiled a RISC-V CPU into Qwen3.5-4B's weights, with no training, and says it runs 18 million instructions without an errorQwenputer is Qwen3.5-4B with a 32-bit RISC-V CPU compiled directly into 11 extra layers, using the model's Gated DeltaNet memory as RAM. Developer Divyansh Agrawal says no weights were trained, the CPU passes the official RISC-V tests, and Qwen can call it as a tool for exact arithmetic. What was built, how, what 'deterministic' means and the limits.14:01·7 min read

Products · ExplainerDoes Claude train on your data? Yes on Free, Pro and Max if Model Improvement is on; no by default on Team, Enterprise and the APIDoes Claude train on your data? On Free, Pro and Max, Anthropic trains on your chats and Claude Code sessions when the Model Improvement setting is on, and keeps that data up to five years. Team, Enterprise and the API aren't used for training by default. Every plan, the setting, the exceptions and how long data is kept.14:01·8 min read

Models · NewsGenie 4: Google DeepMind's world model adds audio and live steering, but it's trusted testers only, with no price yetGenie 4, announced by Google DeepMind on October 27, 2026, generates interactive worlds with sound, dialogue and text-prompt edits in real time. It's a research preview for trusted testers; Google AI Ultra subscribers can join a waitlist for Project Genie, and no price, resolution or frame rate has been published.14:01·6 min read

Policy · NewsOpenAI lawsuit over epilepsy death: parents allege ChatGPT coached their daughter off her seizure medicationAn OpenAI lawsuit over an epilepsy death was filed on October 27, 2026: the parents of Bowen Wan, 31, allege ChatGPT running GPT-4o guided her to taper off her anti-seizure medication and hide it from her doctors. What the suit alleges, the claims, and what OpenAI has said.14:01·5 min read

Policy · NewsBan the term 'artificial intelligence': what Trump posted on October 27, and what the September order already doesTrump says he will ban the term 'artificial intelligence': in an October 27, 2026 Truth Social post, he said the administration would 'in short order' be 'officially BANNING the term.' No new order had followed as of October 29. A September 29 executive order already tells federal agencies to say 'Super Intelligence' instead.14:01·4 min read

Industry · NewsAnthropic government discount: 40% off Claude Enterprise, Claude for Government and the API through GSA's OneGovAnthropic's government discount, announced October 27, 2026, gives agencies 40% off Claude Enterprise, Claude for Government and the Claude API through GSA's OneGov program, including state, local, tribal and territorial governments. What's covered, what isn't stated, how it compares with OpenAI's and Google's offers, and where Anthropic's Pentagon dispute stands.14:01·6 min read

Industry · NewsAnthropic Mithril acquisition: Anthropic buys a one-person muscular dystrophy startup and starts its own drug programsThe Anthropic Mithril acquisition, disclosed on October 27, 2026, brings a one-person rare-disease startup in house as Anthropic starts drug discovery of its own. What Mithril is, what Anthropic's 'early progress' post claims from six drug-development partners, and what's still undisclosed.14:01·6 min read

Industry · AnalysisDoes AI make you work more? What the data says about AI and longer hoursDoes AI make you work more? An October 27, 2026 essay says AI agents left its author working harder than ever. The largest studies find no change in paid hours, but surveys, a time-use study and fieldwork point to more intense, longer days. What each study measured, and what it can't show.14:01·9 min read

Hardware · ExplainerM6 MacBook Pro for AI: what the Dual 16-core Neural Engine, 32GB memory ceiling and 170GB/s bandwidth mean for running local modelsIs the M6 MacBook Pro good for AI? Apple's October 27, 2026 M6 MacBook Pro and iMac top out at 32GB of unified memory and 170GB/s of bandwidth. What Apple claims for AI, what fits in that memory, and when the M5 Pro or M5 Max models still make more sense for local models.14:01·8 min read

Open source · ExplainerWhat is an abliterated model? How refusal removal works, why abliterated DeepSeek is now sold by API, and the safety questionsWhat is an abliterated model? It's an open-weight model whose weights have been edited to strip out its learned refusals, based on 2024 research showing refusal sits along a single direction inside the model. In October 2026, an API router began selling an abliterated DeepSeek V4.1 Flash from five providers. What the technique is, what the licenses and terms allow, and what safety researchers say.14:01·8 min read

Products · ExplainerOffline AI assistant: how Survival LLM pairs a local model with a 930,000-article emergency library, and how people run AI with no internetAn offline AI assistant runs the model and its reference library on your own computer, so it keeps working when the internet doesn't. Survival LLM, which drew wide attention on October 27, 2026, bundles one of four models with about 930,000 Wikipedia articles and medical and repair references. What it needs, how Kiwix, Ollama, llama.cpp and LM Studio work offline, and where offline AI falls short.14:01·9 min read

Developers · AnalysisWikipedia AI-generated content: how editors find it, delete it and now ban it, and what the numbers showWikipedia's fight with AI-generated content has three layers: a volunteer WikiProject started in 2023, a speedy-deletion rule for unreviewed AI articles added in August 2025, and a March 2026 English Wikipedia guideline banning LLM-written article text. How each works, the signs editors look for, and what the Wikimedia Foundation says.14:01·8 min read

Research · AnalysisOpenAI solved open math problems, it says: local smoothing, the Exponential Time Hypothesis and QCD, and what has actually been checkedOpenAI says its AI solved open problems in math, computer science and physics, announced October 26, 2026: Sogge's local smoothing conjecture, a disproof of the Exponential Time Hypothesis, and existence proofs for quantum field theories including QCD. Two come with Lean proofs; the physics work hasn't been released, and no outside group has finished checking any of them.14:01·9 min read

Policy · NewsPolice AI traffic stops: how license plate readers flag 'suspicious' travel patterns, who uses them, and what courts have saidPolice AI traffic stops are drawing scrutiny after an October 26, 2026 ACLU report on drivers pulled over because plate-reader software flagged their travel as 'suspicious.' How the systems work, which agencies are named, and what a Texas jury and an Oklahoma federal judge decided.14:01·6 min read

Developers · NewsMLflow RCE CVE-2026-116692 (CVSS 9.8), Google's MCP Toolbox file-exfiltration flaw and a critical LiteLLM advisory: which versions fix themThe MLflow RCE CVE, CVE-2026-116692, is a CVSS 9.8 flaw in third-party scorer loading, fixed in MLflow 3.17.0. It was published on October 27, 2026, a day after Google's CVE-2026-93328 for MCP Toolbox for Databases and alongside a CVSS 10.0 LiteLLM Enterprise advisory. Affected and fixed versions for all three.14:01·7 min read

Open source · NewsLiquid AI LFM3: six 2B-class open-weight models for text, vision, audio, omni and decisions, free only below $10 million in revenueLiquid AI LFM3, released October 27, 2026, puts text, vision, audio, omni and two 'decision' models on one 1.7B backbone. What 'every modality' means, the sizes, the LFM Open License's $10 million revenue limit, and where the models run.14:01·6 min read

Industry · NewsAmazon AGI lab shut down: what's confirmed about the July 2026 cuts, the San Francisco site closure and the Nova modelsDid Amazon shut down its AGI lab? Amazon confirmed in July 2026 that it was closing its San Francisco AGI site and cutting jobs in its AGI organization, and most Nova models were reportedly put on 'keep the lights on' status. A July Futurism story resurfaced on Hacker News on October 27. What's confirmed, and what Amazon still runs.14:01·6 min read