techWed 7 Oct, 07:30

Sources93 X accounts · techtwitter.com · Hacker News · 8 newsletters
  1. Runs every day at 07:30 London time.
  2. It is shown its own papers of the last 3 days, to know what it has already said.
  3. Haiku 5.5 writes the paper from what was fetched, with no tools.
  4. Jev scores every story by how much it matters to you; each section puts those first, and what is not for you drops to the briefs.

93 X accounts

Their posts of the last 24 hours, through FxTwitter; plus everclear077/awesome-ai-x-accounts when it answers.

techtwitter.com

What tech Twitter is discussing, ranked by discussion.

Hacker News

The front page, plus stories over 80 points in the last 24 hours.

8 newsletters

Every new issue, each delivered once, looking back up to 7 days.

The tech daily written for one reader

Mistral Large 4: Europe's new flagship

Mistral released "Le Chonk," a 1T-parameter open-weight model with 49B active parameters. It's the best open-weight from US or Europe on aggregated benchmarks, beats GPT-6 Astra on visual grounding, and wins on critical workloads like cyber defense and finance. Built end-to-end in Europe, deployable via Mistral Cloud. Weights coming end of October.

OpenAI math GitHub repo

700+ solutions including quasi-Riemann hypothesis work. Significant analytical number theory breakthroughs.

Vibecoding vs hand-coding satisfaction

Some devs saying agents code faster but it's less fun than writing by hand. Counter: that's what happens when your joy came from the mechanical act of turning tickets into code, not from caring about the software itself.

React ecosystem releases: Preact 11, Nuxt 4.6, SvelteKit 3, Remix 3

Preact 11 after 6 years with improved hydration and automatic ref forwarding. Nuxt 4.6 server-agnostic (Nitro optional). SvelteKit 3 cleanup. Remix 3 stable with no React required.

OpenAI and Anthropic: The risk-tolerance divide

Sam Altman says AI's benefits justify accepting some bad outcomes and prefers lighter-touch regulation. Anthropic's position: we must pace the frontier and fears loss of control. This is not a new fight, but Altman is drawing the clearest line yet between the two labs.

Big announcements

Wednesday, October 7, 2026 · Pergent2

OpenAI publishes solutions to hundreds of unsolved math problems

OpenAI released a GitHub repo with solutions to over 700 math problems previously unsolved, including a quasi-Riemann hypothesis proof that researchers call "one of the most important advances in analytical number theory in decades." Also resolved decidability problems. Consulted with Institute for Advanced Study's advisory group on responsible release.

Claude joins Google Workspace

Claude now works directly inside Google Docs, Sheets, and Slides via sidebar. Files also open inside Claude. The interface reads what you have open and edits in place with an approval step before changes land.

Google releases EmbeddingGemma 2

Open, lightweight multimodal embedding model with 740M parameters. Handles text, code, images, audio, and video in a single embedding space. Modular (down to 128 dimensions), matryoshka embeddings for storage, 8K context window, Apache 2.0 license.

OpenAI Decisions API goes public beta

Real-time general classifier makes decisions up to 10x faster than GPT-6 Luna through the Responses API. Lets apps choose the right model, tool, or action near-instantly.

Anthropic expands Cyber Verification Program

Security professionals now get broader access to Claude Mythos 5.1, Opus 5.5, and Sonnet 5.5 with safeguards for defensive work. New tiers open authorized offensive work like penetration testing and red-teaming.

Model Hardware Standard spreading beyond labs

Anthropic's MHS (think "MCP for hardware") got 30x more organization interest than planned in first two weeks. Already in use: QuEra controlling quantum lasers (695/700 success), CMU automating lab experiments with AI agents, UW connecting six lab instruments in a week.

What people are fighting about

Wednesday, October 7, 2026 · Pergent3

Prompting advice just flipped

bcherny: forget the scaffolding tricks and special formatting. Just talk to Claude like you'd talk to a coworker about what you want, how hard it should work, and how to verify it succeeded. The Sonnet 3.5 days of obsessing over prompt structure are gone; models are too capable now.

Meta Muse's creepy relationship mapping

Researchers extracted Muse's internals and found it creates hourly profiles of everyone in your life: facts (where they live, birthdays), history, and suggestions for improving each relationship. Privacy researchers say it's "honestly pretty creepy," but Meta argues all this context is what makes agents useful—and you control what it sees.

LinkedIn drowning in AI slop

Gergely Orosz notes startup blogs are now mostly AI-generated gibberish nobody would've published before agents. LinkedIn added an "AI write" button and saw engagement explode; the platform could stop it but short-term juice wins over long-term health.

Should we use native browser APIs?

Even Nolan Lawson, who usually defends "use the platform," acknowledged native controls are often too limited. Kicked off a huge HN debate on whether it's the browser's fault or unrealistic dev expectations.

Web dev corner

Wednesday, October 7, 2026 · Pergent4

TanStack Charts 1.0: composable charting

Framework-neutral charting library that's typed, tree-shakable, and renders to SVG or Canvas. Samples show line, bar, pie, heatmaps, box plots, histograms. Declarative and composable.

Terminal status standard (OSC 7501)

mitchellh published a spec letting programs tell terminals what they're doing: idle, working, waiting, finished, failed. Designed to help with the "agentic inbox" problem when many agents are running.

From the newsletters

Wednesday, October 7, 2026 · Pergent5

Meta Muse: The privacy cost of agents

The Deep View deep-dives into Muse's relationship-mapping system. Creates hourly profiles on everyone in your life with facts, history, and relationship-improvement suggestions. Meta says context makes agents useful and you control it. Privacy researchers call it creepy. The tension between usefulness and surveillance is the story.

Nolan Lawson: Why don't developers use the platform?

JavaScript Weekly's top story. Even the "use the platform" advocate concedes native controls are often too limited. Kicked off a huge HN thread on whether it's the browser's fault or unrealistic dev expectations.

Research & open source

Wednesday, October 7, 2026 · Pergent6

Worth a read

Wednesday, October 7, 2026 · Pergent7