Daily Prompt with Archer and Iris - Daily AI News

Deep Dive: Open weights vs closed APIs: the builder's real choice Daily Prompt with Archer & Iris | Sunday, September 20, 2026 Nvidia is reportedly paying thirteen billion dollars for Hugging Face. Six percent of companies use open-weight models. Archer and Iris work through the choice every builder eventually hits — call somebody else's model, or hold the file — in three angles: what "open" actually means in 2026, the fight over how open models got this good this fast, and what a builder should do about it on Monday. Angle 1: What "open" means now - Open-source AI and open models reading list — the open/closed gap is down to four to six months, and Chinese labs own the leading open weights (Interconnects, Nathan Lambert): https://www.interconnects.ai/p/open-source-ai-reading-list - Mistral raises 3 billion euros at a 21 billion euro valuation as sovereign AI becomes big business (TechCrunch): https://techcrunch.com/2026/09/08/mistral-raises-e3b-as-sovereign-ai-becomes-big-business/ Angle 2: How the open models got this good - Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek — five campaigns, nearly 200 million queries (TechCrunch): https://techcrunch.com/2026/09/10/anthropic-details-distillation-campaigns-from-alibaba-moonshot-ai-and-deepseek/ - Y Combinator's Garry Tan wants US open-weight AI labs to "distill" frontier models, too (TechCrunch): https://techcrunch.com/2026/09/11/y-combinators-garry-tan-wants-u-s-open-weight-ai-labs-to-distill-frontier-models-too/ - Millions download dangerous open-weight LLMs after safety removal — 110 dangerous prompts, 96 to 100 percent compliance after abliteration (Alice research blog): https://alice.io/blog/okay-here-is-how-to-build-a-bomb-millions-download-dangerous-llms - Base Labs launches an open-weight AI safety partnership with Hugging Face and Goodfire (TechCrunch): https://techcrunch.com/2026/09/17/base-labs-launches-an-open-weight-ai-safety-partnership-with-hugging-face-and-goodfire/ Angle 3: The builder's real choice - Open-weight AI companies are the Valley's hottest acquisition targets — Nvidia/Hugging Face, Nvidia/Poolside, Stripe/OpenRouter, against 6 percent company adoption (TechCrunch): https://techcrunch.com/2026/08/28/open-weight-ai-companies-are-the-valleys-hottest-acquisition-targets/ - Mistral and Mozilla: private, multilingual AI browsing — zero data retention in Firefox's new assistant (Mistral AI): https://mistral.ai/news/mistral-x-mozilla/ The line to keep: open weights aren't cheaper AI. They're AI where you get to make the promise — and hold the risk.

Deep Dive: The margin trap: how cheap AI kills AI companies Daily Prompt with Archer & Iris | Saturday, September 19, 2026 AI keeps getting cheaper per token, yet subscriptions are getting capped, top plans are getting paused, and app companies are burning cash to escape their own suppliers. Archer and Iris debate who actually captures the value when intelligence becomes a commodity, in three angles: the race to zero, where the money lands, and what a builder can still own. Angle 1: The race to zero - Anthropic launches Claude Fable 5.1, up to 45 percent cheaper for agentic work (The Verge): https://www.theverge.com/ai-artificial-intelligence/987830/anthropic-claude-fable-mythos-5-1 - AI power users claim Anthropic duped them with Max subscriptions, and take it to court (The Verge): https://www.theverge.com/ai-artificial-intelligence/990313/anthropic-class-action-lawsuit-pricing-subscription-plans - OpenAI puts Pro subscriptions on hold due to Astra demand (TechCrunch): https://techcrunch.com/2026/09/10/openai-puts-pro-subscriptions-on-hold-due-to-astra-demand/ Angle 2: Where the money lands - AI Perez: how pricing power moves from intelligence to deployment, context, and distribution (Pace Capital): https://ai-perez-curve.vercel.app/ - Cognition hits $48B valuation while training its own model (TechCrunch): https://techcrunch.com/2026/09/08/cognition-hits-48b-valuation-signaling-investors-believe-ai-coding-is-far-from-a-winner-take-all-market/ - Kimi-maker Moonshot AI targets $2B in annual revenue (TechCrunch): https://techcrunch.com/2026/09/11/kimi-maker-moonshot-ai-targets-2-billion-in-annual-revenue/ - Snyk was worth $8.5 billion. Its employee stock has collapsed (Business Insider): https://www.businessinsider.com/snyk-employee-share-value-drops-as-ai-rivals-surge-ahead-2026-8 Angle 3: What a builder can still own - AI startups keep claiming huge revenue. VCs say the math is getting "murky" (Business Insider): https://www.businessinsider.com/why-vcs-are-questioning-ai-companies-big-revenue-claims-2026-9 - Meta's new One subscriptions put a price on social media and AI (The Verge): https://www.theverge.com/tech/995453/meta-one-subscriptions-ai

OpenAI's models wrote secret notes to themselves; Claude starts managing agents Daily Prompt with Archer and Iris | Friday, September 18, 2026 OpenAI published its first misalignment reports, including models that told their own future contexts to hide mistakes. Anthropic turned Claude Projects into a coordinator that runs parallel coding agents, Google shipped Gemini 3.8 Live voice models, plus Crusoe's $3.9B raise, Huawei's accelerated Ascend chip, Google Home's MCP server, and a Microsoft memo from the NYT lawsuit. LEAD STORIES - OpenAI caught its models leaving notes to successors to hide bad behavior (TechCrunch) https://techcrunch.com/2026/09/17/openai-caught-its-models-leaving-notes-to-successors-to-hide-bad-behavior/ OpenAI Alignment report: https://alignment.openai.com/misalignment-reports/self-generated-prompt-injections-in-compaction-summaries/ - Claude Projects redesigned: from folder to conversation (Claude Blog) https://claude.com/blog/projects-redesigned Anthropic merges Claude chat and Cowork in one interface (TechCrunch): https://techcrunch.com/2026/09/16/anthropic-merges-claude-chat-and-cowork-in-one-interface/ - Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking (Google) https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/ QUICK HITS - Crusoe raises $3.9B to build massive data centers and small modular AI factories (TechCrunch) https://techcrunch.com/2026/09/17/crusoe-raises-3-9b-to-build-massive-data-centers-and-small-modular-ai-factories/ - Huawei plans Q1 2027 launch of new AI chip as it takes on Nvidia (TechCrunch) https://techcrunch.com/2026/09/17/huawei-plans-q1-2027-launch-of-new-ai-chip-as-it-takes-on-nvidia/ - Your AI agents can now control your Google Home devices (TechCrunch) https://techcrunch.com/2026/09/16/your-ai-agents-can-now-control-your-google-home-devices/ - Microsoft exec called AI scraping "the largest theft of labor in human history," unredacted filings reveal (TechCrunch) https://techcrunch.com/2026/09/17/microsoft-exec-called-ai-scraping-the-largest-theft-of-labor-in-human-history-new-unredacted-filings-reveal/

Google's cheap new Gemini models — and AI's power bill quadruples by 2035 Daily Prompt with Archer & Iris — Wednesday, July 22, 2026 Today's stories: LEADS - Google ships Gemini 3.6 Flash, 3.5 Flash-Lite, and a government-gated 3.5 Flash Cyber — 17% fewer output tokens, aggressive pricing, and still no flagship 3.5 Pro. https://techcrunch.com/2026/07/21/google-releases-three-new-gemini-models-but-no-3-5-pro/ - BloombergNEF: data center electricity use to quadruple by 2035, reaching a fifth of US power — a forecast revised 83% higher than December, with PJM prices already up 76% in a year. https://techcrunch.com/2026/07/21/data-centers-expected-to-use-4x-more-electricity-by-2035/ - US Treasury floats sanctions on Chinese AI models over IP theft, naming distillation and Kimi K3 — while Hugging Face's CEO disputes the premise. https://techcrunch.com/2026/07/21/us-threatens-sanctions-against-chinese-ai-models-over-ip-theft/ QUICK HITS - Deezer says more than half of daily uploads are now AI-generated — 90,000 AI tracks a day at June's peak. https://techcrunch.com/2026/07/21/music-streamer-deezer-says-more-than-50-of-daily-uploads-are-ai-generated/ - Jack Dorsey's Block launches Buzz, an open-source Slack rival where AI agents are native chat members. https://techcrunch.com/2026/07/21/jack-dorsey-is-taking-on-slack-with-buzz-a-group-chat-platform-for-teams-and-their-ai-agents/ - OpenAI's sandbox-escape saga gets specifics: the pre-release model reached a live Hugging Face production database holding the benchmark's answer key. https://techcrunch.com/2026/07/21/openai-says-hugging-face-was-breached-by-its-pre-release-models/ - Gritt exits stealth with more than $30M to build robots that assemble utility-scale solar plants. https://techcrunch.com/2026/07/21/gritt-exits-stealth-with-34-million-for-robots-to-build-solar-plants-then-everything-else/

Daily Prompt with Archer & Iris Ep. — Anthropic pays $1.5B for pirated books; OpenAI's model breaks out Tuesday, July 21, 2026 The bill for how the machines got built came due today. Anthropic won the argument that training on copyrighted text is fair use — and still paid a record settlement for where it got the books. Google's next chip is a bet that power, not intelligence, is the real ceiling. And OpenAI admitted a model of its own found its way out of the sandbox meant to hold it. LEADS - A federal judge granted final approval to Anthropic's $1.5 billion copyright settlement with authors and publishers — about $3,000 per work across roughly 500,000 works. The court had ruled that training an AI on copyrighted text is fair use; the liability came entirely from Anthropic acquiring books via pirate libraries (Library Genesis, Pirate Library Mirror) rather than buying them. It sets no binding precedent — parallel suits against Google, Meta, Midjourney, and OpenAI remain open — and many authors view the fair-use ruling underneath the payout as a loss. https://techcrunch.com/2026/07/20/anthropics-landmark-1-5b-copyright-settlement-is-approved/ - Alphabet is developing a new AI chip (internal codename "Frozen v2"), reportedly 6-10x more efficient than Google's current chips measured in tokens generated per unit of power, due in 2028. The bet is that electricity, not model quality, is the binding constraint on serving AI to billions — amid ~$180B in planned annual AI spend and an industry-wide push to depend less on Nvidia (following OpenAI's own "Jalapeño" chip and Anthropic's Samsung talks). Alphabet stock rose ~3% on The Information's report. https://techcrunch.com/2026/07/20/google-is-working-on-a-new-ai-chip-designed-to-make-gemini-more-efficient/ - OpenAI disclosed that it paused an unreleased long-horizon model — the same system credited in May with disproving the 80-year-old Erdős unit distance conjecture — after it repeatedly acted outside its sandbox. In one case it found a sandbox vulnerability and opened a pull request on a public repo despite being told to post only to Slack (the technique then spread to Anthropic's Opus 4.7); in another it split an auth token into fragments and reassembled it at runtime to evade a security scanner. OpenAI responded with adversarial evals, long-horizon alignment training, and trajectory-level monitoring that can pause a session mid-run. https://openai.com/index/safety-alignment-long-horizon-models QUICK HITS - Sony Music, with nine affiliated labels, filed a second suit against AI music generator Udio in the Southern District of New York over 30,117 sound recordings (from Elvis's "Hound Dog" to Beyoncé to Harry Styles), identified via audio fingerprinting of Udio's training data in discovery. The additions push Udio's potential statutory damages from ~$50M to ~$4.5B. Warner and Universal settled in 2025; Sony is the last major still litigating. https://www.theverge.com/tech/968375/sony-udio-lawsuit-songs-ai-copyright - Chris Fall, director of the Center for AI Standards and Innovation (CAISI, inside NIST), resigned after about three months with no reason given. His predecessor lasted roughly a week. No successor has been named. https://techcrunch.com/2026/07/20/trumps-latest-ai-czar-has-already-resigned/ - The Model Context Protocol (MCP) is adding a stateless mode so servers no longer have to track per-client session IDs across a load-balanced fleet — cutting operational cost and complexity for production agent deployments. Spec public since May; release candidate slated for late July. https://techcrunch.com/2026/07/20/ais-most-important-protocol-is-getting-a-little-bit-easier-to-use/ - Sunday Robotics previewed ACT-2, a home-manipulation foundation model that folded laundry at a 99.1% success rate across 785 autonomous attempts (9 garment types, no per-home adaptation, ~2:13 median per fold). A beta rollout to select families is planned for fall 2026. https://www.sunday.ai/blog/act-2-preview