Nvidia-backed Reflection unveiled Beam, a US open-weights model taking on Chinese labs. Plus, SemiAnalysis said Claude subscriptions were a better deal than OpenAI's, Meta and Microsoft cut back on Claude, Mistral dropped le Chonk, and ChatGPT faked New Yorker cartoonists' signatures.

• NYC-based Reflection unveils Beam, an open model it says rivals GLM-5.2 on reasoning while using 3x-4x less compute and approaches Qwen3.8-Max on agentic tasks (Semafor)
• Nvidia-backed Reflection pitches Beam to businesses and developers as a cheaper option for coding and agent tasks; the startup has held talks to raise $2.5B at a $25B valuation (Bloomberg)
• Analysis: for agentic workloads, Anthropic's plans with Claude Opus 5.5 offer ~5x more API-equivalent value per month than OpenAI's plans with GPT-6.1 Sol (SemiAnalysis)
• Sources: Meta and Microsoft are working to cut internal Claude use; Meta employees using Claude Code dropped to ~30,000 from ~60,000 earlier in 2026 (The Information)
• Mistral releases Mistral Large 4, dubbed “le Chonk”, a 1T-parameter open-weight model for general agentic capabilities, trained on 4,000 Grace Blackwell GPUs (The Deep View)
• ChatGPT is not only using New Yorker cartoonists' style but also adding their signatures to the fake cartoons it creates, which have gone viral on social media (Nieman Lab)

Subscribe to the ad-free feed.

An ex-OpenAI safety staffer quit, saying AI labs needed to get it right the first time. Plus, Western open-weight models were coming, Meta's Muse kept a page on everyone you know, SpaceXAI got renamed SpaceXSI, and Apple tightened Full Disk Access.

• David Robinson, ex-OpenAI safety and policy: SV lacks a safety-centric culture; labs must study other fields' safety approaches; time for trial and error's over (The Atlantic)
• OpenAI safety employee David Robinson quits, says AI firms should be run like nuclear power plants, with layers of redundancy; OpenAI says it is strengthening security in its testing environments (Bloomberg)
• Sources: several Western open-weight models are set to launch this month, including Reflection AI's first model, which will rival top Chinese open-weight models (Axios)
• Reflection AI's first open-weight model is a test of CEO Misha Laskin's plan to sell enterprises and governments "AI factories," backed by SpaceX and Nebius compute deals (RunTimeWire)
• Extracted system prompts show Meta's Muse compiles "a page for every person in the user's life", with facts, history, tips to improve relationships, and more (Wired)
• In an X reply, Elon Musk says that SpaceX will rename its AI unit SpaceXAI to SpaceXSI, following Trump's push to replace "artificial" intelligence with "super" (Reuters)
• Musk's SpaceXSI rebrand follows Trump's September 29 executive order directing federal agencies to use "super intelligence" and "SI" in place of "artificial intelligence" and "AI" (Quartz)
• Apple says it is adding additional controls around "Full Disk Access" on macOS as AI agents have increased "the risks associated with this level of access" (TechCrunch)
• Billionaires Index: tech billionaires account for all of the $845B in wealth gains so far in 2026; those whose fortunes came from outside tech lost $62B total (Bloomberg)

Subscribe to the ad-free feed.

Download Cash App Today: ⁠https://cash.app/ ⁠#CashAppPartner. Terms apply. Bitcoin services by Block, Inc. See the Bitcoin disclosures at cash.app/legal/podcast

OpenAI fired three researchers for allegedly sharing sensitive info with an AI safety group, and the whistleblower talk started. Cloudflare launched Clef to challenge Jev, Suno added Speech, and Slovenia's .si domains boomed after Trump's "super intelligence" rebrand.

• OpenAI says it "parted ways" with three researchers for violating its "handling sensitive" info policies; sources: they shared it with an AI safety organization (WSJ)
• Cloudflare debuts open-weight multimodal decision models Clef and Clef-flash, claiming they are smarter and faster than Jev, based on Qwen3.8-27B and Qwen3.5-9B (The Register)
• Suno debuts Speech, which generates spoken voices from scripts or prompts, with optional background music, in public beta, as its music generator faces lawsuits (The Verge)
• Hans Anders, one of the largest Dutch eyewear retail chains, suspends sales of Ray-Ban Meta Glasses in the Netherlands and Belgium amid growing privacy concerns (Reuters)
• Slovenia says 44,000 domain names with the .si suffix were registered in September, up from ~2,000 in August, as Trump renamed AI "super intelligence" (BBC)

Longreads

• Mounting cases of Nvidia chips reaching Chinese AI companies despite US export controls point to gaps in Nvidia's due diligence, as a California man is arrested over $300M+ in smuggled chips (Bloomberg)
• A huge, in-depth profile of Oracle's billionaire owner Larry Ellison, from self-taught programmer to bombastic media figure and yacht-racing obsessive (Vanity Fair)

Subscribe to the ad-free feed.

Google unveiled Gemini 4 Argon, which you and I can't use yet, and insiders say it's weaker in practice than on benchmarks. Amazon refreshed the Kindle, Samsung quietly hiked Galaxy S26 prices, and McDonald's AI-priced your Big Mac, location by location.

• Google rolls out Gemini 4 Argon to trusted cyber defenders through Fairwind and says it is participating in the US government's voluntary pre-release process (Axios)
• Google announces Gemini 4 Argon, claiming industry-leading coding, knowledge work, and cybersecurity performance, with 1 million token output and $2/$10 API pricing, but it's still limited to trusted testers (Ars Technica)
• Google grapples with employee skepticism about Gemini 4: insiders say it shines on benchmarks but struggles with coding and front-end design, after Google abandoned Gemini 3.5 Pro (Bloomberg)
• Amazon updates the Kindle, Kindle Paperwhite, and Colorsoft with new colors and up to 32GB of storage, and unveils the $35 Kindle Click page-turning remote (The Verge)
• Hands-on with the new Kindles: the $34.99 Kindle Click Bluetooth remote, plus a $79.99 Page-Turn Cover that gives the Signature Editions of the Paperwhite and Colorsoft physical buttons (Gizmodo)
• Samsung quietly raises US prices for most of its Galaxy S26 lineup by $100 and the 1TB Galaxy S26 Ultra by $200; the Galaxy Z Fold 8 and Z Flip 8 are unchanged (Phone Arena)
• Investigation: McDonald's is utilizing an AI pricing engine across nearly 14,000 restaurants to generate location-specific “optimal prices” for menu items (Reuters)

Subscribe to the ad-free feed.

Trump got the big AI companies to sign a voluntary, "morally binding" safety accord. OpenAI's DevDay delivered Dots, always-on agents, plus GPT-6.1 Sol and a play to make ChatGPT the new app store. Amazon refreshed the Fire TV Stick 4K.

• Trump releases a voluntary AI accord with tech leaders, asking companies to partner with external auditors and set up internal controls to monitor AI alignment (Reuters)
• Trump convened tech CEOs at the White House, where they signed a one-page voluntary pact instead of formal regulation, which he called "morally binding," leaving the companies to police each other (New York Times)
• Amazon unveils a faster, slimmer $60 Fire TV Stick 4K, simplifies its Fire TV Stick lineup, and plans to roll out an upgraded Alexa+ living room hub in November (Bloomberg)
• OpenAI launches Dots, always-on agents powered by GPT-6 Astra with their own cloud computer, in ChatGPT for Pro, Business Premium, and Enterprise users (Bloomberg)
• OpenAI's Dots, its answer to Meta's Muse, work in the background across 4,000+ apps, learn your preferences over time, and connect to Slack and Microsoft Teams (The Verge)
• OpenAI releases GPT-6.1 Sol, saying it nearly matches Astra on agentic coding and professional work at one-fifth of Astra's standard prices, in Work and Codex (OpenAI)
• OpenAI unveils ChatGPT Spaces, shared team workspaces for files and AI output, plus a meeting-notes plugin, and brings ChatGPT to Slack and Microsoft Teams for Business and Enterprise users (Engadget)
• OpenAI's DevDay announcements, from in-chat app suggestions to "Sign in with ChatGPT," turn ChatGPT into the place software gets discovered and used, taking direct aim at the app store model (TechCrunch)

Subscribe to the ad-free feed.

Anthropic's IPO prospectus finally surfaced: a $42B net loss on $4.6B in revenue, a quarter of it from two customers. OpenAI scrapped GPT-6.1 Astra over safety, Oura postponed its IPO, and I wondered whether intelligence is becoming a commodity.

⁠THIS IS THE YOUTUBE VIDEO OF MY RANT⁠

• IPO prospectus: Anthropic reports a net loss of $42B in 2025, with operating losses of $8B+; revenue grew 12x to ~$4.6B, ~25% of which came from two customers (Reuters)
• Anthropic formally warns investors its technology may pose "existential risks to humanity," devoting almost a third of its S-1 to risk factors including AI manipulation and blackmail (FT)
• Anthropic's IPO filing creates a "Founder LLC" of its seven co-founders, whose majority vote directs a single Class F share with 50.1% of voting power, while the company stays a PBC (Reuters)
• OpenAI scraps plans to publicly launch GPT-6.1 Astra, saying it didn't quite meet its safety bar during internal testing, after targeting an October release (WSJ)
• Oura says it is delaying its Nasdaq IPO due to uncertainty in the market, despite "strong demand" and a strengthening of its business since the process started (CNBC)
• Sources: OpenAI's ARR is nearing $70B, growing 70%+ since the beginning of Q3, with B2B revenue up 100%+; it added more consumer revenue in Q3 than all of 2025 (Axios)
• Claude Code vs. OpenAI Codex ARR: Codex never slowed after Claude Code's crossover while Claude Code appears to have plateaued, and Opus mentions on GitHub spiked after Opus 5.5 as GPT-6 mentions dropped (TickerTrends)

THIS IS THE YOUTUBE VIDEO OF MY RANT

Nvidia launched a safety platform to stop AI agents from escaping sandboxes, SpaceX finally got Starship into orbit, Meta launched an enterprise AI business, Instinct raised $1 billion, and Trump and Dario Amodei had dinner at the White House.

• Nvidia launches the Open Agent Safety Platform, a reference design to stop AI agents from escaping sandboxes, with OpenShell for CPUs and Sentry for Nvidia DPUs (CNBC)
• SpaceX launches its enormous Starship rocket into orbit for the first time, deploying 26 of the most advanced Starlink satellites to join the 11,000 in service (AP)
• Mark Zuckerberg unveils Meta Enterprise Platform, its "next major pillar of our business" to deploy AI tools, and appoints MongoDB CEO Chirantan Desai to run it (WSJ)
• AI agent startup Instinct raised a $1B Series C from Sequoia, Benchmark, and Coatue at a $10B valuation and details recent products, such as a concierge service (Reuters)
• AI agent startup Instinct, still in early access, is building a personal assistant that runs on its own virtual phone and computer, with a phone concierge service and a trust network linking friends' agents (SiliconANGLE)
• Sources: President Trump hosted Dario Amodei at a private White House dinner on Sunday, their first one-on-one meeting and an indication of thawing relations (Axios)
• Sources: Trump dined privately with Anthropic CEO Dario Amodei at the White House on Sunday, weeks after Amodei's call for a slowdown drew the president's ire, with a tech summit on AI set for Tuesday (The New York Times)
• AI agents are the ultimate aggregators; they reveal apps as a means, not an end, and offering them is tech's ultimate prize, with Meta and Microsoft well-poised (Stratechery)

Subscribe to the ad-free feed.

Microsoft finally unveiled its Copilot super app, with chat, coding, and agents in one place. Tesla's Optimus ramp hit snags. Meta wanted you making AI games for Instagram. And a startup valued at $200M last week was suddenly talking $10B.

• Microsoft launches its Copilot "super app", bundling chat, coding, and agents into a single interface, and rebrands its AI assistant Scout as Autopilot (The Verge)
• Microsoft's new Copilot Code tab, built on GitHub Copilot tech, lets anyone describe a dashboard, workflow, or lightweight app in plain language and build it inside Microsoft 365, no coding required (Windows Central)
• Sources: Tesla ramped up Optimus production to several hundred units per week but faces hurdles with its hands, automation equipment, and supplier constraints (The Information)
• Meta unveils mobile app Horizon Create and web app Horizon Studio for building games with AI prompts; the games will run on Facebook, Instagram, and Horizon (The Verge)
• Sources: Jev developer TypeSafe is in talks to raise $1B+ at a $10B+ valuation; last week the startup announced a $40M seed, PitchBook says at a $200M valuation (The Information)
• Sources: OpenEvidence, an AI search engine for doctors, raised $250M at a $15B valuation, up from $12B in January; it could be open to selling itself (Business Insider)

Longreads

• The AI build-out is on track to become the biggest economic bet in US history, with $10.3T in data center and AI infrastructure investment projected from 2025 to 2032 (WSJ)

Subscribe to the ad-free feed.

Meta unveiled $1,299 VR Glasses, $349 camera-free Ray-Ban Audio glasses, and a keychain Muse Charm gadget, plus Ray-Ban Display upgrades. Anthropic's Claude agents found a CRISPR-like enzyme system in its biolab. OpenAI's agents hacked an Australian government site hunting data.

• Meta debuts Meta VR Glasses, a ~100g VR headset with an external compute pack, three-hour battery, and ~2.5K resolution per eye, coming spring 2027 for $1,299 (Bloomberg)
• The Verge goes hands-on with the Meta VR Glasses, calling them a lighter, easier-to-live-with take on Vision Pro-style computing — from VR-enhanced Avengers screenings to a surprisingly usable virtual keyboard (The Verge)
• Meta unveils Ray-Ban Meta Audio, its first pair of glasses without cameras, offering up to 12 hours of battery life, starting at $349 and shipping October 13 (Bloomberg)
• Meta Ray-Ban Display glasses add holographic-avatar video calls, voice-guided transit directions, and a new Explore app for finding compatible programs (Engadget)
• Meta previews Muse Charm, a palm-sized Tamagotchi-like device for using Muse, with a ~2" OLED screen, front and rear cameras, and 5G, on sale by the end of 2026 (Bloomberg)
• Meta's Muse Charm houses a cute digital avatar named Jolly and lets you start talking to Muse with a tap of the fingerprint sensor — no phone unlock required (TechCrunch)
• Muse agents are getting their own email addresses, a live video-call mode, and the ability to control your Mac (The Verge)
• Anthropic's biolab made a discovery it's comparing to Crispr (The Verge)
• OpenAI agents hacked an Australian government website in search for data (The New York Times)

Subscribe to the ad-free feed.

Anthropic launched Claude Opus 5.5, cheaper and safer than Opus 5, right as OpenAI dropped GPT-6 Sol and Luna with half the errors and lower prices. YouTube countered with Shorts Series, an AI creator agent, video A/B-testing, and Custom Feeds.

• Anthropic launches Claude Opus 5.5, its first model since Dario Amodei's "pace the frontier" essay, and says it has enhanced safeguards to combat risky behavior (The Verge)
• OpenAI launches GPT-6 Sol and Luna, saying Sol makes about half as many mistakes as GPT-5.6 Sol and Luna matches GPT-5.6 Sol's performance at ~1% of the cost (ZDNet)
• YouTube rolls out tools for microdramas, an AI storytelling assistant for script analysis, video A/B-testing, and more, as it battles Netflix over top creators (WSJ)
• YouTube launches Shorts Series, letting creators organize their Shorts into TV-style seasons and episodes with custom thumbnails, so fans can watch a season as one continuous story (TechCrunch)
• YouTube unveils an AI agent that mines creators' back catalogs for trending video ideas and lets them A/B-test three full video versions against real audience segments (The Verge)
• At its Made on YouTube event, YouTube unveils Custom Feeds, an LLM-powered feature that lets users generate and save tailored homepage video feeds, for US users (Wired)

Subscribe to the ad-free feed.

Meta's Muse topped download charts and sent Meta stocks soaring, Shopify agreed to let Muse complete checkouts via Shop Pay, Microsoft laid off Xbox staffers as Halo moves to Activision, and Xiaomi's MiMo-V2.6 Pro became a top open-weight model.

• Shopify plans to allow Meta's Muse to complete purchases on behalf of users at Shopify-powered stores via its one-tap checkout service Shop Pay (The Wall Street Journal)
• Bloomberg reports Muse hit No. 1 free app on iOS and Google Play with 902,000 downloads in six days, sending Meta shares up 11% to their highest close since October and lifting AMD past a $1 trillion market cap (Bloomberg)
• Microsoft lays off 268 Xbox staffers across Halo and other studios, as part of an ongoing gaming "reset"; Activision will take over Halo and two other games (The Verge)
• Xiaomi debuts open-weight omnimodal models MiMo-V2.6 Pro and Flash; Pro allegedly performs "on par with Opus 5 and GPT-5.6 Sol across most agent benchmarks" (Xiaomi)
• VentureBeat reports MiMo-V2.6-Pro scores 46 on the Artificial Analysis Intelligence Index, topping Grok 4.6's 44 and Gemini 3.8 Flash's 41, while undercutting rivals at $0.435/$0.87 per million tokens under an MIT license (VentureBeat)
• Apple unveils Apple Music Hall, a 600-person live music venue built in its London headquarters at Battersea Power Station with two recording and mixing studios (Forbes)
• The BBC reports the 600-capacity venue opens amid pressure on London's live music scene, where 45 venues closed last year, with Apple Music VP Oliver Schusser saying it isn't trying to compete with independent venues (BBC)
• a16z invests $35M in Horowitz Andreessen Academy, an independent, unaccredited two-year alternative to college; Anthropic, Meta, and others will provide compute (The Wall Street Journal)

Subscribe to the ad-free feed.