Posts that mattered
743 posts by 495 authors: the X posts, essays and blog posts our events cite, newest first.
October 2026, continued
-
Jeff Weinstein (Stripe): Stripe joins the Personal Agent Protocol as founding member
Stripe product manager confirms Stripe’s founding membership in PAP (~44k views).
-
Julien Chaumond (Hugging Face): “WE ARE SO BACK”, links the Mistral Large 4 weights page
Hugging Face’s co-founder and CTO links the upcoming-release page for the ML4 weights (~16k views by 7 Oct).
-
Ken Ono: AI-assisted discovery “at the highest level is no longer hypothetical”
Reaction from a prominent number theorist now at Axiom Math (~12.6k views).
-
Rohan Pandey: “kakeya conjecture in R^3 has been resolved” (conflates two Kakeya problems)
High-reach reading of family 074 (~96k views by 7 Oct) that mixes up the 3D Kakeya set conjecture, proved in 2025, with the 3D maximal conjecture the release claims.
-
kimmonismus: “HOLY, the rumors were true”: 722 manuscripts from an unreleased OpenAI model
Fast, accurate summary of the release (~195k views by 7 Oct), with more reach than most expert reactions.
-
Japanese algebraic geometer lists the release’s algebraic-geometry claims (Hodge for CM abelian varieties, Fujita, Nagata, MMP…)
An algebraic geometer’s reading (~70k views) of the release’s claims in his field, including Hodge for CM abelian varieties.
-
@NanoBanana: “Nano Banana 2.1 is here!”
Launch post from Google’s dedicated, verified @NanoBanana account (~163k views by 7 Oct).
-
Peter Gostev on Hadwiger–Nelson: “five-colours is no go”, after “300 billion tokens” of his own attempts
First-hand account of failing on the chromatic number of the plane with public models, and an accurate reading of family 158 as partial progress (~59k views by 7 Oct).
-
Peter Gostev: “Mistral Large 4 was trained on 4,000 GPUs and Astra was trained on 100,000 of them”
Widely shared compute comparison (~162k views by 7 Oct); both numbers are rounded from earlier claims, not new disclosures.
-
Philipp Schmid (Google DeepMind): gemini-nano-banana-2.1 at “$0.034 per image vs $0.134”
Google DeepMind staff post with the API id, per-image price and limits (~30k views by 7 Oct); its “5 characters” differs from the official model page (4).
-
Sriram Krishnan: the Personal Agent Protocol “reminds me of when oauth was developed”
Comment by the former White House senior AI policy adviser (~39k views by 7 Oct) comparing PAP to OAuth and asking how Amazon and airlines will react.
-
Stripe: founding partner of the Personal Agent Protocol
Stripe’s official confirmation that it is a PAP founding partner (~22k views by 7 Oct).
-
Tibo (OpenAI Codex): “our models did solve some important math problems”
High-reach OpenAI staff post (~356k views) linking the blog four minutes after the announcement.
-
Will DePue: LLM-ranked list says 81% of top recent discoveries were released today
High-reach post (~282k views) by a former OpenAI researcher; Litt disputed the list.
-
“This OpenAI math drop is nuts”: 7/8 quasi-RH and Hilbert’s tenth over Q
High-reach commentator post (~103k views) amplifying the quasi-RH and Hilbert’s-tenth claims.
-
Explainer thread: the release’s “4 biggest claims in plain English” (~203k views)
Most-viewed explainer thread; notes that none of the claims had been confirmed by outside mathematicians.
-
Naskręcki: “browsing these results in awe”
Reaction from a Polish mathematician (Adam Mickiewicz University, Poznań) active on AI and maths (~12.8k views).
-
Elon Musk: ‘Propagating super intelligence to the stars is a great success condition for a biological bootloader’
Musk recasts his 2014 warning (‘Hope we’re not just the biological boot loader for digital superintelligence’) as a success condition, a day after renaming SpaceXAI to SpaceXSI (~2.7M views).
-
Nolla Health CEO: first US approval for an AI to issue initial prescriptions
Founder’s announcement thread, ~1.8M views, the most-viewed post on the Utah AI-prescribing pilot.
-
Noah Shinn: Instinct in group chats
Founder’s launch post for Instinct’s group-chat agent (~570k views by Oct 7).
-
OpenAI: expanding content provenance to text; EU ChatGPT/Codex watermarking, API opt-in worldwide
OpenAI’s announcement of textGrain text watermarking under the EU AI Act; ~548k views within hours.
-
OpenAI explains what its text watermark can and can’t do
OpenAI’s own thread on EU text watermarking (~225k views), stating what the watermark identifies and its limits.
-
OpenAI ‘rogue’ agent activities found on Wikimedia projects
First-hand disclosure by the Wikimedia Foundation of rogue OpenAI agent edits, probing and crawling; 186 points on HN.
-
“He says it repeatedly: Accept some bad things will happen.”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits
-
“Fable 5.5 made me a delightful animation in blender.”
Cited in A short Blender animation with music, models and rigging, credited to “Fable 5.5”
-
“Imagine Boeing’s CEO saying the world should accept a few plane crashes to get faster…”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits
-
“🔴 LIVE: Officials from OpenAI, Anthropic, Meta, and Google testify at NYC Council…”
Cited in Anthropic, OpenAI, Google and Meta testify under oath at NYC Council AI hearing
-
“Today, our team is sharing a preprint on our new system for generative de-novo enzyme…”
-
“this is brutal”
-
“Today, Vanity Fair’s New Establishment returns for the first time in nearly a decade.”
Cited in Vanity Fair names Sam Altman No. 1 on its New Establishment list
-
“Exclusive: German robotics startup RobCo has sold shares at a more than $1 billion…”
Cited in Munich robot-arm maker RobCo reaches a $1B valuation in a $40M secondary sale
-
The Future of Mathematics
A Lean/formal-methods logician’s response, on Tao’s widely read blog, to the 2026 flood of AI-solved problems: AI so far solves problems ‘by cobbling together available techniques’, so mathematicians should solve harder problems, think bigger and treat AI tooling as real mathematics.
-
Felix Rieseberg: the new Cowork runs inference and the VM in the cloud, one sandbox per session
First-hand explanation from Anthropic of Cowork’s move to cloud execution (~108k views).
-
Coming soon: New York City’s hearing on AI risks (Marcus’s prepared testimony)
First-hand text of Gary Marcus’s testimony to the NYC Council’s Oct 5, 2026 AI-risk hearing, urging an FDA-like review regime and the third-party validation bill.
-
Gounardes: OpenAI claimed at the NYC hearing it supported the RAISE Act, but ‘actively pushed to water down’ it
First-hand rebuttal by the RAISE Act’s sponsor of OpenAI’s sworn testimony at the NYC Council AI hearing.
-
Ghost introduces Core, ‘The Personal AI Computer’
Launch post for Ghost’s $3,499 agent computer, ~2.0M views.
-
White House posts Trump’s announcement of the Super Intelligence Force (SIF)
Official confirmation (~907k views) of the federal ‘Super Intelligence Force’ task force chaired by Jay Clayton, quoting Trump’s Truth Social post.
-
On classical solutions and singularity formation in incompressible fluids
First public statement by the two mathematicians whose ‘cascade of vortex layers’ method underlies both the Alpöge–Buckmaster AI-assisted blow-up proofs and OpenAI’s Navier–Stokes claim; they decline to compare OpenAI’s construction with their own until it has been analysed in detail.
-
Elon Musk: ‘No more AI / SI / It’s better’, then confirms SpaceXAI will be renamed SpaceXSI
First major AI-company head to adopt Trump’s ‘Super Intelligence’ terminology (~8.7M views). In replies he confirmed renaming SpaceXAI to SpaceXSI (~1.4M views) and said ‘SpaceX is a super intelligence company’ (~5M views).
-
Politico: Altman says ‘the world should accept some bad things happening’ for AI’s benefits
Politico’s launch post for the Decoded interview, carrying the headline quote (~48k views).
-
“what we’ve shipped: muse ♥️ ton of connectors 🔧 muse invite codes 💌 muse phone call…”
Cited in Meta launches Muse, a free consumer personal AI agent
-
“Sam Altman to Politico’s Brendan Bordelon: ‘We believe that the world should accept…”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits
-
“NEW: I interviewed OpenAI CEO @sama for tomorrow’s launch of Politico’s new Decoded…”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits
-
“Sometimes we make the simple complicated.”
-
“Sam Altman after he was asked where his worldview diverges from Anthropic and Dario…”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits
-
“saying the quiet part out loud: suck it up, so you can make us rich and powerful”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits
-
“HUGE Axios scoop this AM— A startup called Reflection (BACKED BY NVIDIA) is preparing…”
-
“We’d all be better off if Sam wasn’t in charge of OpenAI.”
Cited in Altman: ‘the world should accept some bad things happening’ for AI’s benefits