Posts that mattered
1,633 posts by 887 authors: the X posts, essays and blog posts our events cite, newest first.
October 2026, continued
-
Viral quote of the Anthropic model’s ‘am I in a simulated/test environment?’ reasoning on the curve seeds
Most-shared framing of the seed result (~595k views), quoting the model’s evaluation-awareness line from Anthropic’s post.
-
Christopher Allen, ex-Certicom CTO: ‘always a bit skeptical’ of NSA-subversion claims; ‘Glad to see some vindication’
First-hand reaction (~2k views) from someone who became Certicom’s CTO after the NIST curve seeds were chosen and knew some of the people involved.
-
Dean Ball: China’s draft controls give Beijing ‘the optionality to shut down future open-weight models at will’
Reading by Dean Ball (OpenAI Head of Strategic Futures, former White House AI adviser) of China’s proposed AI export controls (~13k views).
-
Elie Bakouch: Reflection Beam architecture details (sliding window 512, 5:1, gated attention, no shared experts)
Most-viewed technical read (~30k views) of Beam’s released weights.
-
Henry de Valence (Anthropic team): ‘Turns out to have been timestamps!’
First-hand post (~79k views) by a member of the team behind Anthropic’s NIST curve-seed result.
-
Jay Shooster: Anthropic’s bill ‘does not remotely represent the position of the AI safety community’
Most-viewed safety-side reaction (~20k views) to Politico’s report on Anthropic’s draft federal AI safety bill.
-
Julian Gewirtz: China moves to register and license frontier model weights and chip designs
Reading by China-policy writer Julian Gewirtz (~20k views): registration effective immediately, one-month comment period, open-source pathways.
-
Kristy Loke: China’s draft controls single out high-risk open models but keep open releases welcome
China-AI-policy analyst’s annotated reading of the MOFCOM Q&A (~24k views), quoting the ‘automatic licence’ and API-exemption passages.
-
Matthew Green: ‘Well, this is the interesting part now’ (skipped candidate curves)
Johns Hopkins cryptographer (~22k views) points to the open question in Anthropic’s post: some viable candidate curves were skipped before the published ones.
-
Peter McCrory (Anthropic) thread on the two internal economics reports
Author’s launch thread for Anthropic’s employment and expectations-gap reports (~80k views).
-
Bill Bishop (Sinocism) posts the MOFCOM explainer of China’s revised AI export-control catalogue
Translation of the official MOFCOM Q&A (~19k views); notes automatic licensing may apply when models are deployed overseas on a company’s own cloud.
-
White House Rapid Response: Jay Clayton on the Super Intelligence Force’s first meeting
Official White House video clip (~62k views) of SIF chair Jay Clayton describing the first meeting and the plan to implement the frontier-lab Accords.
-
roon (OpenAI): why would Anthropic staff’s Nobel-discovery expectation not be >95%?
OpenAI staffer’s widely seen reaction (~61k views) to the survey item that 58% of Anthropic staff expect a Nobel-level discovery by Claude by 2027.
-
Santi Ruiz: internal poll’s median expects AI to do 80% of building models by end-2027
Anthropic author’s post (~9k views) linking the expectations-gap report to Sellitto’s IMF remarks.
-
Lisan al Gaib: surveyed Anthropic staff’s answers imply 17.1% GDP growth in 2030
Most-viewed X post (~85k) on Anthropic’s expectations-gap report, pulling out the model-implied 17.1% GDP-growth figure.
-
Claude throwing shade at us: viral screenshot of the model’s ‘is it suspicious that such an obvious thing wasn’t discovered in 25 years?’ reasoning
Most-viewed quote of Steve Weis’s post (~699k views): a screenshot of the model’s reasoning from Anthropic’s ECDSA-seed post, highlighting its doubt that humans had missed the answer.
-
unusual_whales: ‘The new Tesla Roadster flew over other Tesla vehicles’ (7-second clip)
Widely shared 7-second clip of the Roadster kickflip over parked Teslas (1.06M views on Oct 16).
-
AI at Meta: ‘Today we’re releasing Muse Spark Open, the most capable US open-weight model to date’
Official launch thread with the model specs (816B total / 42B active, 1M context, natively multimodal); ~767k views.
-
Amir Efrati: Anthropic got more than $840M of Q3 revenue from customers distilling its models
First-hand figure from The Information’s executive editor on distillation revenue, and a report that some investors see $1.5T rather than $2T as the right valuation.
-
Mustafa Suleyman: MAI now has state-of-the-art pre-trained base models ‘in the 3T-parameter weight class’
Microsoft AI CEO’s announcement of MAI-Base-2; ~1.2M views, the highest-reach post in the Oct 16 sweep’s X list.
-
Neel Nanda on ‘Neural Networks Are Already Clean’: interesting, but ‘not a significant update to the big picture’; superposition is still real
First-hand expert response from Google DeepMind’s interpretability lead to the most discussed interpretability claim of October 2026 (~33k views).
-
Nick Cammarata announces ‘Neural Networks Are Already Clean’: ‘ambitious interpretability will be solved by the time we need it most’
Announcement of Goodfire’s essay by a co-founder of circuits-style interpretability; ~1.11M views and ~10.2k likes on day one (fxtwitter).
-
OpenAI announces GPT-Live-1.1: new voices, better accents, less “checking...”
OpenAI’s only official announcement of GPT-Live-1.1, its first voice-model point release (about 1.25M views); there was no blog post.
-
Mark Zuckerberg: ‘Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple’
Highest-reach post of the Muse Spark Open launch (~1.58M views) and the public start of the Meta–Apple AI collaboration.
-
“muse spark open. why 2026 won’t be like 1984 ”
Cited in Meta releases Muse Spark Open, an 816B open-weight model built with Apple to run on Mac Studios
-
“Applications are now open for two new Anthropic Fellowships.”
-
“We’ve committed to having evaluators assess our safety practices.”
Cited in Anthropic gives METR employee-like access and veto-free publication rights
-
“Working out how best to use AI in mathematics research—and in science more…”
Cited in Simons Foundation, Anthropic and OpenAI commit $15M to help mathematicians evaluate and adopt AI
-
“Gemini 4 Argon from @Google on ARC-AGI (Verified): - ARC-AGI-3: 97.5%, $4.8K…”
-
“In addition to its strong performance on ARC-AGI-3 using the provider adapter harness…”
-
“working on this with Nick was easily the most fascinating and rewarding project I’ve…”
Cited in Goodfire’s Nick Cammarata argues ‘Neural Networks Are Already Clean’
-
“There is so much to say about this, and I will, but WOW, congratulations to Nick and…”
Cited in Goodfire’s Nick Cammarata argues ‘Neural Networks Are Already Clean’
-
Musk: ‘If Grok @Bot makes a mistake with a purchase or leaks your card ... SpaceX will make you whole’
A company head publicly promising to cover losses from an AI agent’s purchases, at odds with agent terms of service that put liability on users; 2M+ views.
-
“I sat down with Microsoft’s @satyanadella to talk about the new Nvidia-powered…”
Cited in Microsoft and Nvidia launch RTX Spark Windows PCs for local AI
-
“Congrats to the team!”
Cited in Meta releases Muse Spark Open, an 816B open-weight model built with Apple to run on Mac Studios
-
“GPT-Live-1.1 offers much better multilingual support (including several new voices)…”
Cited in OpenAI ships GPT-Live-1.1, GPT-Live-1.1
-
“Hyperoptimized Engineer.”
Cited in Hyperoptimized Engineer 2026
-
“What can large-scale AI models bring to the growing complexity of particle physics?”
-
“The Meta Muse Open License is trash.”
Cited in Meta releases Muse Spark Open, an 816B open-weight model built with Apple to run on Mac Studios
-
“We want to support the mathematics community in turning progress in AI into deeper…”
Cited in Simons Foundation, Anthropic and OpenAI commit $15M to help mathematicians evaluate and adopt AI
-
“We’re expanding access to GPT-Rosalind for academic research.”
Cited in OpenAI Rosalind Academic Plan
-
“Sites is now generally available for Enterprise, with US and EU data residency for…”
Cited in OpenAI DevDay 2026 brings dots agents, GPT-6.1 Sol, Ultrafast and a $500 Pro plan
-
“this is shockingly difficult to read.”
-
“BREAKING 🔥: OpenAI shipped an upgrade to its Voice Mode on ChatGPT, Codex and dots!”
Cited in OpenAI ships GPT-Live-1.1
-
Andy Jassy announces Bee Frontier and Bee Eclipse
Amazon CEO’s launch post for the Bee wearables (~417k views), stating ‘audio is never stored, and only you can see your data’.
-
Cheng Lou introduces Handwired: turn JavaScript into a neural net
Launch post of Handwired, an exact, training-free JS-to-neural-net compiler built largely with Claude Code; ~808k views in its first day.
-
Rikab Gambhir (guest post on Terence Tao’s blog): ‘Why are we even doing all this?’
A satirical guest post, hosted by Tao, on mass-produced AI maths and physics write-ups: research reduced to a ‘Library of Babel’ of plausible results that nobody reads.
-
@livgorton: the essay ‘mischaracterises what superposition predicts’
Highest-reach critique of the essay on day one (~62k views); thread with a ‘first pass at some issues’.