Posts that mattered
1,633 posts by 887 authors: the X posts, essays and blog posts our events cite, newest first.
October 2026, continued
-
Alex Heath: Sources interview with Qualcomm CEO Cristiano Amon on the ‘AI phone supercycle’
Primary source: the interviewer’s post of the full Amon interview, with timestamps and links.
-
“I’ve just made a live tracker for the community effort for optimizing the ‘Integer…”
-
“motion designers are cooked i asked Haiku5.5 to show me how good it is at motion…”
Cited in Haiku 5.5 makes a motion-design showreel with code and Blender in 24 minutes
-
Cloudflare Radar: Yandex Cloud traffic about 33% below the previous week after the Sasovo strike
First-hand network measurement of the first Yandex data-center strike.
-
“Reposting because this is similar to a weaker improvement of OAI’s result discussed…”
Cited in OpenAI: elliptic curves over imaginary quadratic fields are modular
-
“Oh jk got my wires crossed.”
-
“So cool! The improvements in the end started getting faster than we could run the…”
-
“I can’t take Sonnet’s crab seriously with those eyes 😂🦀 Claude Opus 5.5 vs Claude…”
Cited in Fluffy-crab same-prompt comparison
-
“Claude Haiku 5.5 实测来了,Claude Haiku 5.5(OpenCode) 对比 GPT-6 Luna(Codex)的鹈鹕测试…”
Cited in SVG cycling-pelican comparison
-
Jeremiah Fowler: ‘russian bot traffic has dropped to 49%’ after the Yandex data-center strike
Most-viewed X post on the Yandex data-center strikes (2.75M views); an unverified reading of Cloudflare data.
-
Viral reading: ‘Google just announced that Gemini is not a model, it’s actually the agent harness, and they are going to add Claude models’
Most-viewed third-party reading of the Gemini agent launch (~398k views); it captures the rebranding of ‘Gemini’ as an agent that can run Claude.
-
Google Cloud: ‘our new Gemini agent for your business’, a universal cloud agent running 24/7
Official Google Cloud announcement post for the Gemini agent; about 470k views (fxtwitter, Oct 9).
-
Lisa Abramowicz: FT story on OpenAI’s lower revenue sends Oracle down 5%+, showing how leveraged tech is to OpenAI and Anthropic
Bloomberg anchor’s market reading of the FT story (~141k views on Oct 10, fxtwitter).
-
Google: ‘Gemini is becoming your universal agent for work’ (#GeminiAtWork)
Google’s official summary of the Gemini agent launch; ~57k views (fxtwitter, Oct 10).
-
Google’s Shubham Saboo: ‘10 things that make it a universal work agent’ — the Gemini agent can run Gemini 4 Argon or Claude Opus 5.5
Detailed feature list of the Gemini agent from a Google AI product manager (~120k views on Oct 10, fxtwitter): coworker agents with their own Workspace accounts, four memory layers, model choice including Claude, hard spend caps.
-
Swapnil Jain: OpenAI problem #109 tightened to κ = 2^-27, building on Colkitt
Earliest X update we found in Swapnil Jain’s parallel Claude-assisted series; it explicitly builds on Doug Colkitt’s results.
-
Sundar Pichai on the Gemini agent at Gemini at Work (NASA Hangar One, 700+ customers)
CEO post on the Gemini agent launch; about 319k views (fxtwitter, Oct 9).
-
Pirat_Nation: Astrocade reaches 10 million monthly users for AI-made games
High-reach summary of the Astrocade growth numbers cited in the entry (88k views).
-
Claude: ‘Introducing Claude Haiku 5.5’, about 75% cheaper to run than Haiku 4.5
Official launch post for Claude Haiku 5.5; about 6.24M views on Oct 9 (fxtwitter), far more than any other post on the launch.
-
Justin Drake: crypto should plan for ‘bunker mode’ as AI may break ECDSA ‘in months not years’
5.5M-view call by a leading Ethereum researcher to migrate funds to fresh addresses, citing OpenAI’s math release as ‘mathematical superintelligence’.
-
Elon Musk: Grok Bot will use ‘the best back end model for any given task’, including Claude Opus 5.5, Midjourney and Suno
SpaceX’s owner says its own agent product will route work to rival companies’ models, including Anthropic’s Claude Opus 5.5 (~21M views in five hours). A Grok Bot team member then said the Bots themselves will run on Opus 5.5.
-
“OpenAI cannot make AI safe on its own”: open letter from the three fired OpenAI safety researchers
Primary account from the fired researchers disputing OpenAI’s misconduct claims and asking labs to stop work that reduces chain-of-thought monitorability.
-
Altman: “We are entering a new era of discovery now”
OpenAI CEO’s framing of the math release; ~520k views.
-
Vitalik Buterin: don’t scramble, but take AI-accelerated math risks to cryptography (incl. lattices) seriously
Ethereum co-founder’s 2.2M-view reply names lattices (ML-DSA, FHE) as the new AI-vulnerable area and argues for hash-based cryptography and 10x key sizes.
-
Doug Colkitt: OpenAI problem #109 tightened to κ = 2^-59 (from 2^-182)
Third conditional tightening in one day (κ = 2^-59, a 2^123-fold larger exponent saving than OpenAI’s original), the post that went viral through quote posts.
-
Doug Colkitt: OpenAI problem #109 tightened further to κ = 2^-78
Highest-reach post of the series (~135k views): second conditional tightening, to κ = 2^-78, via nonadjacent axis swaps.
-
Doug Colkitt: conditional improvement of OpenAI problem #109 (integer multiplication), κ from 2^-182 to 5.8×10^-33
First announcement (02:24 UTC, ~105k views) of a conditional tightening of the exponent in OpenAI’s claimed sub-n log n integer multiplication algorithm.
-
Artificial Analysis: Claude Haiku 5.5 scores 43 on its Intelligence Index, uses ~3x GPT-6 Luna’s tokens at max effort
First independent benchmark read-out of Haiku 5.5 (index, token use, hallucination rate, AutomationBench refusal issue); about 385k views (fxtwitter, Oct 9).
-
‘Skate 3 is now FULLY PLAYABLE in the browser!’: most-viewed post of the AI-decompiled game-port wave (~3.9M views)
Highest-reach post found on the wave of browser game ports that Kotaku attributes to AI decompilation (~3.86M views by Oct 11), though the post itself does not mention AI.
-
Claude: ‘One more thing’, cache reads on Sonnet 5.5 halved to $0.10 per million tokens
Official announcement of the Sonnet 5.5 cache-read price cut that came with Haiku 5.5; about 1.54M views (fxtwitter, Oct 9).
-
ClaudeDevs: monthly Claude Platform API credits for Max ($100/$200) and Team (up to $500) plans
Primary announcement of the monthly API credits bundled with subscriptions; about 3.2M views on Oct 9 (fxtwitter).
-
Elon Musk: ‘We will build and run the fab’; TSMC may at most sublease part of Terafab
Musk’s own clarification (~340k views on Oct 7) that Tesla and SpaceX will run Terafab themselves. It rules out the TSMC-owned-and-operated scenario that Culpium had called most likely.
-
Independent re-run of the Lean check for OpenAI’s ζ(s) ≠ 0 for Re(s) > 7/8 proof: “the proof checks”
First public independent check of a headline claim: two kernels accept the Lean proof of the 7/8 zero-free half-plane.
-
WSJ’s Max Zeff: fired OpenAI researchers write to board members on monitorability
Scoop announcing the three fired OpenAI safety researchers’ letter urging labs not to decrease AI monitorability.
-
“Progress on BSD is pure PR, no content.”
Cited in OpenAI claims BSD in Selmer corank ≤ 1 and Goldfeld’s conjecture
-
Rybin on OpenAI’s proof that the plane is not 5-colourable: ‘totally unbelievable alien math?’
High-reach (~261k views) first reaction to OpenAI’s claimed lower bound 6 for the chromatic number of the plane, pointing to its transfer to ‘weakly measurable’ colourings.
-
“My famous night train benchmark is in!!! 🚂🌙🔥 Haiku 5.5 vs.”
Cited in Night-train counterexample
-
“just confirming - is this routing based on whatever connectors you have available?”
Cited in Musk: SpaceX’s Grok Bot will use ‘the best back end model for any given task’
-
“Pedi para uma Fable 5.5 contar a história da humanidade em rabiscos 16 eras.”
Cited in Human history in doodles
-
“GPT‑6 with Intelligent UI rolls out globally to Plus, Pro, Business, and Enterprise…”
-
“The twist estimate in OpenAI’s new proof of Polyakov’s conjecture (215), never uses…”
Cited in 2D O(n) mass gap (Polyakov) and O(3) continuum limit claimed (OpenAI)
-
“🚨Haiku 5.5 vs Sonnet 5.5🚨 Same prompt.”
Cited in Electric-motor explainer comparison
-
“Grok goes open. Holy shit.”
Cited in Musk: SpaceX’s Grok Bot will use ‘the best back end model for any given task’
-
“We used Flux for image generation in the early days of Grok.”
Cited in Musk: SpaceX’s Grok Bot will use ‘the best back end model for any given task’
-
“Polyakov’s conjecture on the O(N) mode has been demonstrated by OpenAI.”
Cited in 2D O(n) mass gap (Polyakov) and O(3) continuum limit claimed (OpenAI)
-
“And it was known to be true from numerics for decades”
Cited in 2D O(n) mass gap (Polyakov) and O(3) continuum limit claimed (OpenAI)
-
Brett Adcock: Hark’s first 24 hours (143k daily active users, 1.5M messages)
First-hand launch metrics for Hark Pro from Hark’s founder.
-
Viral claim: AI has solved “90 of the 500 most important open problems” (LLM-ranked list)
Spread the ‘90 of top 500’ figure (~117k views); based on Lech Mazur’s match against an LLM-ranked list.