AI social digest — 2026-09-12
Top line: The beat was the essay's echo chamber: Altman agreed on-platform within hours and Roose christened the "Sam and Dario agree" moment — while the anti-doom counter-wave ran its own escalation, with the Coxon smear campaign graduating to named-funder "psyop" allegations and Hugging Face's leadership joining the pile-on. The discourse is now fighting about pacing machinery by name.
The pacing handshake, X-first (continuation — risk-governance)
What happened: Amodei announced the essay on X (~10:32 AM ET). Altman's agreement landed ~9:30 AM ET — "I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access…" Kevin Roose: "We have reached the 'Sam and Dario agree on something' part of the singularity." "Dario is right" quote-posts followed.
What's genuinely new: Altman's post is a specific, matching commitment — "independent evaluators with employee-like access" is Anthropic's step 1, adopted in public and in real time, not a vibe of agreement. That's the first convergence artifact between the two labs on the record — it was the day's second-biggest post.
Links: https://x.com/kevinroose (RT) · Altman post via feed capture, 2026-09-12 17:30 UTC.
The counter-wave: Coxon "psyop" gets named funders; Hugging Face leadership joins (continuation — risk-governance, discourse strand)
What happened: The dismissal campaign around Jacob Coxon's Sep 9 warning escalated on three axes the same day the pacing essay published:
- Named-funder conspiracy claim: Kevin Bass — "The Coxon psyop was a decade and more than a billion dollars in the making… Coefficient Giving (formerly Open Philanthropy) is the grant-providing vehicle of Dustin Moskovitz…" — i.e., from character attack to an alleged funded operation. Will Chamberlain: "not the first coordinated AI doomer op," citing a Guardian piece.
- Hugging Face leadership pile-on: CEO Clem Delangue — "asking Jacob about AI extinction risk is like asking your AC guy about climate change" — and CTO Julien Chaumond — "if you do believe in non-null existential risk, you should [not] be working at a lab." Notable provenance: HF is the July agent-breach scene, and its two leaders are now in the character-defense chorus.
- Pacing as the explicit target: Daniel Jeffries ran a four-post barrage — mocking data centers as "susceptible to buckets of water attacks," rejecting "doom" economics outright, and pre-emptively attacking any "Oversight Committee for Pacing Economic Innovation and Development" as stacked with "the drinking buddies of the folks stumping for it." Richard Sutton: "too much intelligence is not one of [the world's problems]." Steven Sinofsky and Pessimists Archive ran the résumé/tech-panic framing.
What's genuinely new: (a) the conspiracy allegation now names specific funders — a legal-adjacent escalation that will either produce receipts or defamation exposure; (b) HF leadership's participation; (c) the discourse openly fighting over pacing machinery, not just doom — the essay converted a safety-feud into a governance-feud within hours. Still zero actual reporting on Coxon himself, day 3.
Links: feed captures 2026-09-12 12:30–17:33 UTC (handles above).
Melanie Mitchell: "the last several weeks of AI hell" (new, adjacent)
What happened: AI researcher Melanie Mitchell published a synthesis thread retroactively framing the season's incident pile-up ("I wrote down my thoughts about the last several weeks of AI hell").
What's new: the skeptics' side now has its own retrospective document to match the labs' essays and the July employee statement — the discourse has moved from replies to synthesis artifacts. Worth reading when she posts the full text.
Grok ships into Microsoft 365 Copilot (continuation — openai-bundling)
What happened: Microsoft 365: "Launching even more models in Copilot. Grok models from SpaceXAI are rolling out in Copilot in Word, Excel, and PowerPoint." Nadella: "More model choice coming to Copilot. Welcome Grok!"
What's new vs the ai digest: PowerPoint confirmed in the rollout scope. Same-day counterpoint to the pacing call stands.
Quick hits
- AlphaGenome Atlas (Hassabis): "With AlphaFold we mapped the protein universe — now with AlphaGenome Atlas we're charting the human genome… predict the impact of all 9 billion possible single-letter DNA variants." Vendor claim, no third-party read; filed, not crowned. (2026-09-12 16:31 UTC.)
- Astra's Factorio run: community post — GPT-6 Astra beat Factorio with enemies enabled after ~44h in-game (4d11h on the Codex /goal clock); benchmark-of-one sentiment, and the cost-claim math in the post is unverified. (gpt6-astra color.)
- Grok Bot GTM engine: Yale AI Association demo + student app contest (DogeDesigner RT); the Sep 15–17 Galaxy livestream now has named participants "speedrunning building a company using only Grok Bot" in 3 days. Spectacle-marketing; watch what ships. (agent-harnesses thread.)
- Deutsch on measurement: David Deutsch — "economists and their theories are outdated to measure the benefits of AI." Sentiment, not news.
- SF mood datapoint: "The entire city of San Francisco is spooked and on edge due to AI safety shit" — the discourse now includes ambient-anxiety color. Anecdote, not data.
Dropped
- The Musk-RT sludge layer (Boring Co Nashville tunneling — noted as Tennessee local color only; Tesla Semi/FSD/Solar posts; Terafab nostalgia RT; Honduras Starlink), Bukele education posts ×2, Balestriero's Harvard JEPA talk, OpenClaw client-app RT (community tooling), Grok personal-utility plug, Justine Moore's "Mathematicians" meme, parenting meme, game promos (Master of Orion 2, tldraw), Coxon RTs already captured above.