Posts that mattered
1,633 posts by 887 authors: the X posts, essays and blog posts our events cite, newest first.
October 2026, continued
-
“Over 400 academics in 30+ countries signed our open letter calling for immediate…”
-
Paweł Huryn: Muse Wonder Full fixed 44 of 105 real bugs, one behind GPT-6 Astra and ahead of Fable 5.1 and Opus 5.5
An independent, real-repository test of Muse Wonder Full on the day after launch (about 42k views): it places Meta’s model level with OpenAI’s and Anthropic’s best, in contrast to early ‘benchmaxxed’ complaints.
-
“So it turns out that OpenAI *can* give specific reasons why they fired someone, when…”
-
“Gemini 4 Argon is here. I got early access (thank you friends at Deepmind!)”
-
“In short: Fable 5.5 is the best model in the world, and we’re releasing it in the…”
-
“Asked Gemini 4 Argon for an interactive museum of Apple computers.”
Cited in Gemini 4 Argon launch day on X
-
Aditya Ramabadran quotes OpenAI’s “We used our models to understand the device”
Quote of OpenAI’s line “We used our models to understand the device”; the most viewed reaction to the arrest from a small account (1.9k followers). About 431k views (fxtwitter, 14 Oct 2026 evening).
-
Chetaslua: Gemini 4 Argon sculpts a pagoda temple complex in Blender from Antigravity
Most viewed launch-day Argon demo in the sweep list (Blender temple via multi-agent visual critique loops). About 159k views (fxtwitter, 14 Oct 2026 evening).
-
Daniel Kokotajlo: ‘...they are becoming more situationally aware...’ (Argon report excerpt)
The most viewed post on the Argon safety report’s situational-awareness findings (~198k views).
-
Elon Musk replies ‘Yikes’ to Chamath Palihapitiya’s AI private-credit post
One-word reply (~406k views) cited by Bloomberg and Yahoo Finance in coverage of CoreWeave’s bond and stock fall on Oct 14.
-
fofr: Gemini 4 Argon makes a pelican-on-a-bike video with Fourier epicycles and Lyria 3.5 music
The most viewed Argon launch-day demo in the sweep list (about 244k views on 14 Oct); video made by Argon with music it prompted from Lyria 3.5.
-
Google Gemini app: Gemini 4 Argon is rolling out now in the Gemini app
Official consumer-product announcement of Argon in the Gemini app (thread); about 60k views about 2.5 hours after posting (fxtwitter, Oct 13).
-
Gergely Orosz on the OpenAI bugging arrest: a $1M+ career thrown away, more suspicion for all employees
High-reach reaction on the employee-trust consequences of the OpenAI bugging case. About 129k views (fxtwitter, 14 Oct 2026 evening).
-
J Rosser: CoT controllability evals for Gemini 4 Argon (how far it can steer its reasoning to evade CoT monitors)
First-hand thread on a safety evaluation of Argon done with collaborators @NeelNanda5 and @zimmerrol (per the post); low reach but primary safety information.
-
Keerthana Gopalakrishnan: Gemini 4 Argon ‘will have massive impact on robotics’, with a developer blog
First-hand post from Google DeepMind robotics linking the ‘Gemini 4 Argon for robotics’ developer blog.
-
Surya: Gemini 4 Argon thread of 534 paintings made with code
Small account (7.3k followers) whose launch-day thread of 534 code-made paintings became one of the most viewed Argon demos. About 150k views (fxtwitter, 14 Oct 2026 evening).
-
Koray Kavukcuoglu: launching Gemini 4 Argon for the hardest coding and knowledge work tasks
Post by the author of the launch blog and Google’s Chief AI Architect (low reach); about 10k views about 2.5 hours after posting (fxtwitter, Oct 13).
-
Neel Nanda: the Gemini 4 Argon safety report is ‘Google’s most thorough by far’
Google DeepMind’s mechanistic-interpretability lead points readers to the loss-of-control, alignment and monitorability sections of the 217-page Argon Safety & Security Report (~83k views).
-
Guillermo Rauch: “Never bet against Google” on Gemini 4 Argon
Vercel CEO’s high-reach launch reaction; argues for model-agnostic gateways after Muse Wonder and Argon. About 153k views (fxtwitter, 14 Oct 2026 evening).
-
Susan Zhang: “doomers must be so disappointed it was this guy and not a chinese spy”
High-reach reaction to the OpenAI bugging arrest, mocking expectations of a Chinese spy. About 174k views (fxtwitter, 14 Oct 2026 evening).
-
Zuckerberg: “Introducing Muse Wonder, our new family of frontier models”
Meta CEO’s launch post for Muse Wonder, with by far the widest reach of the launch.
-
Cheeky Pint: ‘Muse Wonder: Zuckerberg and Wang’s $100B AI’ (interview transcript)
The only long first-hand account of how Meta built Muse Wonder: codenames, team size, training cost, the $100–200B compute bet made before the models existed, and what Meta plans next.
-
Alexandr Wang: “introducing muse wonder ... rivals the big bois”
Launch post by Meta’s chief AI officer, who leads Meta Superintelligence Labs.
-
Artificial Analysis: Muse Wonder Full scores 56 and Muse Wonder 52 on the Intelligence Index v4.3
First independent evaluation of the Muse Wonder models, with pricing, token use and cost per task.
-
Zuckerberg: Muse Wonder “brings us to the frontier”; two models released
First-hand statement of what the two models are: Muse Wonder for ‘fast frontier performance’, Muse Wonder Full ‘the full version of the model’.
-
Satya Nadella: ‘our biggest bet yet on our own silicon’ (Maia 300, multi-gigawatt Maia deployments)
Microsoft’s CEO announces Maia 300 and the plan to make the Maia family the primary inference system for Microsoft’s own products from next calendar year.
-
Tim Sweeney: ‘a huge pivot underway here at Epic’ to retool for maximum AI acceleration
Epic’s CEO confirms, hours after Bloomberg’s report, that the company is reorganizing developers around AI-accelerated development.
-
Sophie Huiberts (guest post on Terence Tao’s blog): ‘This is normal bad work’
A mathematician’s argument, hosted by Tao, that OpenAI’s Oct 6 release of 722 manuscripts is ordinary progress written up badly, not a watershed: the papers would be rejected for poor attribution and exposition.
-
Why aren’t AI companies being charged for hacking?
Top AI story on Hacker News on Oct 12, 2026 (314 points): a programmer’s argument that the 2026 intrusions by lab agents show criminal intent by design, not unforeseeable accidents.
-
‘AI’ is Not a Tool
Rich Hickey, the creator of Clojure, argues that LLMs are automation and not tools; 244 points on Hacker News and a 301-point reply the same day.
-
Alexandr Wang: “MUSE WONDER 🍉”: Active Memory, Active Focus, domain expertise
Wang’s own three highlights; the watermelon emoji matches the reported codename ‘Watermelon’.
-
Alexandr Wang: ‘muse wonder over a cheeky pint’
First-hand: Meta’s chief AI officer shares the two-hour Cheeky Pint interview he and Zuckerberg gave on launch day.
-
amit: Zuckerberg on writing ‘this check for ... $100B or $200B’ before seeing the models
Spread Zuckerberg’s Cheeky Pint remark on committing $100–200 billion before Meta had seen its frontier models.
-
Andrew Curran: “Pacing is out the window” after Muse Wonder
Widely read argument that Meta’s frontier release breaks any voluntary pacing among US labs.
-
Andrew Curran: Anthropic’s Frontier Red Team on small models ‘months away’ from self-replication
Widely read summary (about 205k views) that Anthropic’s Logan Graham and Bill Kristol both quoted.
-
Anthropic: ‘significant progress in reducing the effectiveness of distillation attacks against Claude’
Anthropic’s own announcement of its distillation-defenses post; about 1.17M views by midday Oct 13 (fxtwitter).
-
Artificial Analysis thread (reply 1): Muse Wonder Full top three on six of ten Intelligence Index evaluations
Independent per-benchmark results for Muse Wonder Full from Artificial Analysis’ launch-day thread.
-
Artificial Analysis thread (reply 2): Muse Wonder costs $4.52 per Intelligence Index task, 2.8x Muse Spark 1.3
Independent cost and token-use measurements for Muse Wonder.
-
Artificial Analysis thread (reply 3): knowledge ‘a relative weakness’ for Muse Wonder, with low hallucination rates
Independent knowledge and hallucination results for both Muse Wonder models.
-
BridgeMind: ‘Muse Wonder is benchmaxxed’
Most-viewed critical test of Muse Wonder on launch day (about 258k views); Elon Musk replied to it.
-
Chamath Palihapitiya on Maia 300: ‘everyone competes with everyone’
High-reach investor reading of Microsoft’s chip plan (about 335k views); Elon Musk replied to it.
-
John Collison: Cheeky Pint episode with Zuckerberg and Wang on Muse Wonder
Announces a two-hour interview with Zuckerberg and Wang recorded for the launch (chapters on how frontier models are built, ‘Betting $100B’, MSL).
-
David Singleton (Meta Superintelligence Labs): Muse Wonder ‘rolled out first to subscribers earlier today’
First-hand from a Meta Superintelligence Labs executive: Muse Wonder reached Muse subscribers on launch day.
-
Elon Musk replying to Chamath on AI constraints: power, construction, cooling, networking, batteries
Musk’s answer to what limits AI build-out after Microsoft’s Maia 300 news (about 189k views).
-
Elon Musk on Hock Tan’s casino remark: ‘No crying in the casino’
Highest-reach reaction to the Broadcom CEO’s warning about AI investors (about 1.27M views).
-
“You can use Muse Wonder to power your Muse by subscribing today.”
Cited in Meta releases Muse Wonder and Muse Wonder Full, its first frontier-class models
-
Fireside Alpha: Wang says Meta CFO Susan Li warned him ‘if this doesn’t work, this is going to be really, really bad’
Widely viewed clip and transcript from the Cheeky Pint interview: Wang on the financial stakes of Meta’s frontier-model bet.
-
Logan Graham (Anthropic Frontier Red Team head): ‘We think this one is ~pretty underpriced’
First-hand comment from the head of the team that published the self-replication study; names the authors’ X accounts.