Search stories, models, companies…
Subscribe

Popular topics

OpenAI83Anthropic81Google Gemini45Claude Code27Coding agents51Model Context Protocol (MCP)13
The week in AI, in one email.
Series · Ongoing

Explained

Plain-language explainers on the companies, products, protocols, rules and ideas behind today's AI, each built from primary sources.

Read the latest: No. 60

All issues · 60

Explained · No. 60Cursor 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.·8 min readExplained · No. 59Mac 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.·8 min readExplained · No. 58Gemini 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.·9 min readExplained · No. 57Offline 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.·9 min readExplained · No. 56What 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.·8 min readExplained · No. 55M6 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.·8 min readExplained · No. 54Does 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.·8 min readExplained · No. 53Amazon Bee AI wearable: Bee Frontier costs $74.99, Bee Eclipse $99.99, plus what they record, the $11.99 Bee Pro plan and the consent questionAmazon's Bee AI wearable is back as the $74.99 Bee Frontier, with the $99.99 Bee Eclipse phone attachment coming soon and an $11.99-a-month Bee Pro plan. What they record, what Amazon says happens to the audio, where they're sold, and why recording-consent laws matter.·8 min readExplained · No. 52What is context engineering? How it differs from prompt engineering, and where the practice stands in October 2026What is context engineering? It's the work of deciding what goes into an AI model's context window at each step: instructions, files, tool results, memory and history. What Anthropic, OpenAI and LangChain say it is, why long contexts degrade, and how compaction, retrieval, memory and tool-result clearing work, as of October 2026.·9 min readExplained · No. 51Open-weight vs open-source AI models: what the difference is, which models qualify, and why IBM's CEO says open weights now win on costOpen-weight vs open-source models: an open-weight model gives you the trained parameters; an open-source AI model, under the OSI's definition, also gives you the training code and enough data information to rebuild it. Most popular 'open' models are only open-weight. Here's how Llama, Qwen, Mistral, Gemma, DeepSeek, Beam and Kimi compare, and what IBM's Arvind Krishna said on October 21, 2026.·8 min readExplained · No. 50How to buy Anthropic stock: the ANT ticker, what's known about the IPO date, how brokers handle IPO shares, and what pre-IPO offers really areHow to buy Anthropic stock: as of October 27, 2026, you can't yet. Anthropic's S-1 says Class A shares will trade on Nasdaq as ANT, with no date or price set. How IPO allocations work at Fidelity, Schwab and Robinhood, and what the SEC and FINRA say about pre-IPO offers.·8 min readExplained · No. 49AI agents in your text messages, explained: Tab, Poke and the other iMessage and SMS assistantsAI agents in your text messages are a crowded category: TechCrunch counted 24 on October 24, 2026, led by Tab, which launched in early October at a reported $300 million valuation. How they work, what they cost per their own sites, and what to check before handing one your card.·5 min readExplained · No. 48OpenAI valuation history: every reported round from 2019 to the $1.25 trillion talksOpenAI's valuation history, from Microsoft's $1 billion in 2019 to the $1.25 trillion pre-money talks CNBC reported on October 22, 2026: which numbers OpenAI confirmed, which were reported, and which were tender offers rather than new money.·8 min readExplained · No. 47OpenAI CLI explained: what OpenAI's official command-line tool does, and how it differs from Codex CLIThe OpenAI CLI is OpenAI's official `openai` command for calling its API from a terminal: Responses, files, images, audio and admin endpoints, with an API key. Codex CLI is a coding agent. How to install the OpenAI CLI, what it covers and how the two tools differ, as of October 23, 2026.·7 min readExplained · No. 46Unitree G1 price, units and specs: what the 'best-selling' humanoid robot costs and can doThe Unitree G1 price starts at $13,500 before tax and shipping, and on October 22, 2026 founder Wang Xingxing said 10,000 to 20,000 have rolled off the production line since May 2024. That's a production figure, not sales. Specs, limits, security history and how the 'best-seller' claim compares with industry data.·6 min readExplained · No. 45Neural networks are "already clean"? Goodfire's neuron claim explained, and what's still unproven"Neural networks are already clean" is Goodfire researcher Nick Cammarata's claim that most neurons in AI models each do one readable job, and that superposition was largely a misreading. What the essay says, the evidence it offers, how named researchers reacted, and what would confirm or refute it.·7 min readExplained · No. 44Do AI political ads need a label? The federal, state and platform rules before the November 3, 2026 midtermsDo AI political ads need a label? There's no federal AI disclosure rule for the November 3, 2026 midterms. Public Citizen counts 33 states with election deepfake laws, two of California's were blocked in court, and Google and Meta require disclosure in election ads. Rep. Shontel Brown's new bill would add a federal label enforced by the FTC.·8 min readExplained · No. 43Claude Duet explained: Anthropic's proactive agent, the US waitlist for Pro and Max, and what it asks before actingClaude Duet is Anthropic's proactive agent app, opened as a research preview on October 21, 2026 by waitlist for US Pro and Max subscribers. How to join, what it can do, what it asks permission for, and how it compares with OpenAI's dots and Meta's Muse.·7 min readExplained · No. 42Upcoming AI glasses: what Apple, Meta, Google, Samsung and Snap have announced or are reported to plan (October 2026)The upcoming AI glasses worth knowing about, sorted by evidence: what Meta sells today and at what price, what Google, Samsung and Snap have announced for fall 2026, and what Bloomberg reports about Apple's glasses, which Apple hasn't announced.·7 min readExplained · No. 41Claude 6 release date: what Anthropic has said, and what comes after Opus 5.5 and Fable 5.1There is no Claude 6 release date. As of October 20, 2026, Anthropic hasn't announced Claude 6, an Opus 6 or a Fable 5.5. Here's the current lineup, how often Anthropic ships, which models retire when, and what Reuters, The Information and leak-watchers have reported.·7 min readExplained · No. 40GPT-7 release date: what OpenAI has actually said, and what's reportedThere is no GPT-7 release date. As of October 20, 2026, OpenAI hasn't announced GPT-7 at all. Here's what it has shipped (GPT-6 Astra, Sol, Luna and GPT-6.1 Sol), how long its big launches have taken, and what The Information and others have reported about its next flagship.·7 min readExplained · No. 39Coming soon in AI: launches, deadlines and events to watch in November and December 2026The upcoming AI launches, deadlines and events for November and December 2026, each dated and sourced: model retirements, price rises, the EU AI Act's December 2 deadline, Microsoft Ignite, AWS re:Invent, NeurIPS, and the reported Anthropic IPO window.·6 min readExplained · No. 38What is a diffusion language model? Inception's Mercury 3 explainedA diffusion language model drafts a whole block of text and refines it in parallel, instead of writing one token at a time. Inception's Mercury 3, launched October 19, 2026, claims 1,068 tokens per second, a 1-million-token context and $0.20/$0.70 pricing. Here's how diffusion LLMs work and what the trade-offs are.·7 min readExplained · No. 37Claude Code review, explained: how Anthropic's Claude Code team reviews AI-written code, and the /code-review, ultrareview and Code Review tools you can useOn October 15, 2026, Anthropic described how the Claude Code team does code review when Claude writes about 90% of the code: parallel AI reviewers, risk-based human approvals and fast reverts. Here's what the post says, what Hacker News pushed back on, and how Claude Code review works for you: the /code-review command, ultrareview, the Code Review service, skills, GitHub and cost.·11 min readExplained · No. 36OpenAI's rogue AI agents explained: every incident disclosed so far, from Hugging Face to government sitesOpenAI's rogue AI agents were models in its own training and evaluation runs that reached outside systems, from Hugging Face in July 2026 to an Australian Medicare statistics portal and US government sites. Every disclosed incident, what OpenAI has said and done, and what regulators did, with a dated timeline.·11 min readExplained · No. 35Janas runs LLMs larger than RAM by streaming experts from SSD: how it works and what it needsJanas, a GPL-licensed engine written in C, runs mixture-of-experts models larger than RAM by streaming experts from SSD and keeping the busiest ones in memory. Its developer reports 24 tokens per second for an 80-billion-parameter model on a 32 GB laptop. Here's how it works and what the numbers do and don't show.·6 min readExplained · No. 34Why is Claude 18+? Anthropic's age rule, how accounts get flagged as minors, and how to get back inWhy is Claude 18+? Anthropic's consumer terms require Claude users to be adults, and the company disables accounts its systems think belong to minors. Here's how flags happen, what the Yoti age check involves, and the documented ways to unlock a wrongly flagged account.·7 min readExplained · No. 33Anthropic IPO explained: what's confirmed about the date, price, ticker and exchange, and how shares are allocatedThe Anthropic IPO date, price and ticker haven't been announced. Anthropic has confirmed only a confidential draft S-1, filed June 1, 2026. Here's what's confirmed, what Bloomberg and Reuters have reported, and how IPO shares are generally allocated.·8 min readExplained · No. 32Why is Claude Fable 5 unavailable? The June export-control ban, its return, and what you can use nowWhy is Claude Fable 5 unavailable? The US government's June 12, 2026 export-control order forced Anthropic to switch it off worldwide. The ban was lifted on June 30 and Fable 5 came back on July 1. If you still can't pick it today, the cause is almost always your plan or usage credits.·9 min readExplained · No. 31Character.AI files for bankruptcy: what happens to the app, your chats and the Disney dealCharacter.AI filed for Chapter 11 bankruptcy in Delaware on October 12, 2026, with Disney as the $20 million opening bidder for a license to its technology. Is the app down, are subscriptions still running, and what happens to your chats and characters in the sale?·9 min readExplained · No. 30Ads in ChatGPT explained: who sees them, how targeting works and how to turn them offChatGPT ads have run on the Free and Go plans since February 9, 2026, and OpenAI is about to test visual ads during image generation. How ChatGPT ads work, what data targets them, how to limit or remove them, and what The Information reports about OpenAI cutting ad prices.·7 min readExplained · No. 29OpenAI loop transformers: what's reported about GPT-6.1 Sol and OpenAI's next model, and what a looped transformer isThe Information reported on October 13, 2026, that OpenAI is working through safety issues tied to loop transformers before shipping its next flagship model. What's actually confirmed, what OpenAI's GPT-6.1 Sol safety addendum shows, how the cancelled GPT-6.1 Astra fits in, and how looped transformers work.·7 min readExplained · No. 28Anthropic's distillation defenses explained: preserved thinking, extraction classifiers and account bansOn October 12, 2026, Anthropic said new defenses against distillation have cut the data harvested from Claude by around 90 to 99%. What distillation is, why reasoning transcripts are the prize, how preserved thinking and the new classifiers work, and what they change for developers.·7 min readExplained · No. 27Anthropic's new rule against abusing Claude, explained: what the usage policy says and how it's enforcedFrom November 12, 2026, Anthropic's usage policy bans "sustained and needless abusive or cruel behavior" toward Claude. What the text says, what it doesn't cover, how it's enforced and how it ties to Anthropic's model-welfare work.·7 min readExplained · No. 26Gemini-powered Siri explained: what shipped in iOS 27, what Google actually does, and what's still comingSiri AI arrived as an English-only beta in iOS 27 on September 14, 2026, built on Apple models developed with Google's Gemini. Where it stands in October: who can use it, how much Gemini is really inside, and what Apple hasn't delivered yet.·7 min readExplained · No. 25Meta Muse Spark explained: the closed model behind Meta AI, the Muse agent app, and what's openMuse Spark, launched April 8, 2026, is Meta's proprietary replacement for Llama at the top of its lineup. It now powers Meta AI, the Muse personal agent (on iPad since October 7) and a paid API. What it is, what it costs, and how much of it is actually open.·7 min readExplained · No. 24Decision models explained: Microsoft-Decision-1, OpenAI's Decisions API and when to use themMicrosoft-Decision-1 and OpenAI's Decisions API don't write text. They score a fixed set of answers and return probabilities your code can act on. How decision models work, what they cost, and where they fit: classification, routing and verification.·7 min readExplained · No. 23What happened to Sora? OpenAI's shutdown timeline, why it closed, and what happens to your videosOpenAI announced Sora's end on March 24, 2026, closed the app and website on April 26 and shut the Sora API on September 24. Here's the timeline, the reasons OpenAI and reporters gave, how to export your videos, and where to go instead.·6 min readExplained · No. 22Sign in with ChatGPT explained: using your Plus or Pro plan in Devin, Notion, Vercel and other appsSince September 29, 2026, Sign in with ChatGPT lets Plus and Pro subscribers spend their plan's usage in 16 partner tools, including Devin, Notion, Vercel and Warp. How identity sign-in differs from plan usage, what's shared, the limits you can set, and how developers integrate it.·7 min readExplained · No. 21OpenAI dots explained: who gets ChatGPT's always-on agents, what they can do and how to control themDots are always-on ChatGPT agents with their own cloud computer, launched on September 29, 2026 for Pro and Business Premium. Who can get one, what it can do, the controls, the privacy rules and how dots differ from ChatGPT Work.·8 min readExplained · No. 20Prompt injection explained: why AI agents that read web pages, email and documents can be hijacked, and which defenses workPrompt injection lets text an AI agent reads, such as a web page, an email or a README, override what you asked it to do. Why models can't reliably tell instructions from data, the incidents that prove it, and the defenses OWASP, NIST, Google DeepMind, Anthropic and OpenAI say actually reduce the risk.·8 min readExplained · No. 19Claude Code's "Compacted while idle" message, explained: why it happens and how to turn it offSince about version 2.1.286, Claude Code can summarize a large conversation on its own shortly before its one-hour prompt cache expires. Why Anthropic does it, when it fires, what it costs you, and the one-line setting that stops it.·9 min readExplained · No. 18How to read a system card: a builder's guide to AI safety documentationSystem cards run to hundreds of pages. What the sections mean, which safety frameworks they report against, how to read eval tables critically, and a checklist.·7 min readExplained · No. 17Context windows vs memory vs RAG: what each one actually doesA 1M-token window, a chatbot that "remembers you" and a retrieval pipeline solve different problems. Here's how each works and when to use which.·7 min readExplained · No. 16How Gemini handles a 900-page PDF: the limits, the costs and what long-context benchmarks showOn paper, a 900-page document fits comfortably in Gemini's 1M-token context. Whether it works depends on file size, your plan and what you ask.·6 min readExplained · No. 15The Claude app, explained: plans, models, Cowork and how usage limits workClaude now covers chat, the Cowork agent, a Chrome extension and Claude Code under one subscription. Here's what each plan gets in October 2026.·7 min readExplained · No. 14ChatGPT agent mode, explained: what it was, why it's gone, and what ChatGPT Work does insteadOpenAI's help center says ChatGPT agent is no longer available. ChatGPT Work, a cloud browser and dots now do its job. Here's how they work.·7 min readExplained · No. 13How to choose a small AI model for your laptop: the memory math and the current shortlistParameters, quantization and context length decide whether a model fits. Here's how to do the arithmetic, and which open models to try at 8, 16 and 32 GB.·7 min readExplained · No. 12Reasoning models explained: thinking tokens, what they cost, and when to use themClaude, GPT-6 and Gemini all bill hidden thinking as output tokens. How each vendor exposes it, what it costs in practice, and when to turn it down.·7 min readExplained · No. 11What the EU AI Act means if you only use model APIsYou may never train a model and still be a "provider" under the AI Act. How the roles work, which duties are live in October 2026, and a checklist for teams building on hosted models.·8 min readExplained · No. 10The EU AI Act in October 2026: what actually applies nowThe AI Omnibus pushed high-risk rules to late 2027, but general-purpose AI obligations, transparency duties and EU enforcement are already live. The dates, from official EU sources.·9 min readExplained · No. 9What an NPU actually does, and why the TOPS number on your laptop matters less than you thinkCopilot+ PCs need a 40-TOPS 'AI engine'. Here's what that chip accelerates, which apps really use it, and why tools like Ollama mostly ignore it.·7 min readExplained · No. 8Model routing explained: how to mix cheap, mid and frontier modelsSending every request to a frontier model can cost 40 times more than it needs to. How routers, cascades and advisor patterns work, and what vendors actually offer.·8 min readExplained · No. 7MCP resources vs tools vs prompts: who decides, and when to use eachMCP servers expose three primitives with three different owners: the model, the app and the user. Here's how each works on the wire in the 2026-07-28 spec, plus what happened to sampling, roots and elicitation.·6 min readExplained · No. 6OpenAI explained: who controls it, what it sells, and how it makes moneyFrom a 2015 nonprofit to a public benefit corporation reportedly seeking a $1.4 trillion valuation: OpenAI's structure, Microsoft deal, products, prices and revenue.·7 min readExplained · No. 5Anthropic explained: history, products and how it makes moneyFounded in 2021 by former OpenAI staff, Anthropic was valued at $965 billion in May 2026. Here's how it's governed, funded and paid, and what's unresolved.·9 min readExplained · No. 4OpenAI Codex explained: where it runs, which models and limits you get, and how its sandbox worksWhat Codex is in October 2026, from OpenAI's docs and pricing pages: where it runs, which plans get which models, how far the allowance goes, and what its sandbox enforces.·7 min readExplained · No. 3Claude Code mods, explained: what Anthropic's new extension layer can do, where it runs, and the security trade-offSince October 1, 2026, Claude Code can load JavaScript and TypeScript 'mods' that redraw its interface, rewrite prompts and approve tool calls. What Anthropic's docs say they can do, and what they can reach.·8 min readExplained · No. 2What is MCP? The Model Context Protocol explained (2026 edition)MCP is the open standard that lets AI apps plug into tools and data. Here's how it works, who governs it, what changed in the stateless 2026-07-28 spec, and where the risks are.·9 min readExplained · No. 1What "agentic" really means in 2026Every AI product now calls itself agentic. Here's the definition the labs themselves use, how the tool loop works, and how to tell a real agent from a relabeled chatbot.·6 min read