Tickline News

AI social digest — 2026-09-09

2026-09-09 AI social

Top line: X spent the day processing the same two stories the wire filed — Christiano's board seat (welcomed by sama) and the doomer backlash LeCun kept amplifying — plus the math community's Navier–Stokes counter-positioning, Perplexity wiring its index into Nous's Hermes agent, and one fresh item: Perplexity's Q2D-Web retrieval benchmark.

Christiano to the OpenAI Foundation board, and the doomer discourse on X (continuation)

What happened: OpenAI's X account announced Paul Christiano joining the Foundation Board and Safety & Security Committee (1:39 PM ET); sama's welcome: "Welcome, Paul. Grateful you are doing this, and all you have done for AI safety. Excited to work together again" (3:57 PM ET). The day's counter-frame, same platform: LeCun RTing Daniel Jeffries — "AI doomers are delusionally dangerous. Their solutions to an imaginary problem are: 1) Rig[ging]..." (6:57 AM ET) — the same morning The Verge carried Anthropic's >10% extinction warning.

What's actually new: the shape of the safety establishment's response. OpenAI's answer to a week of agent-incident coverage is a governance hire of the alignment establishment's most credible figure; the doomer-bad-faith counter-discourse runs through exactly the people who were least persuaded by the DseWiki evidence. Both are sentiment datapoints; neither is a fact claim.

Links: - OpenAI announcement: https://x.com/OpenAI/status/2097741659509584091 - sama welcome: https://x.com/sama/status/2097776310940569783 - LeCun doomer RT: https://x.com/ylecun/status/2097716899031175579

Navier–Stokes from the mathematicians' side (continuation)

What happened: The math community's X response to OpenAI's claim, via LeCun's RTs: Jitendra Malik — "in view of OpenAI's claim about Navier Stokes, it is important that we hear alternative points of view" (6:34 PM ET Sep 8); Julia Kempe — "Congratulations to Tristan Buckmaster and Levent Alpöge for developing the AI assisted technique" (8:12 PM ET Sep 8); Terry Tao held up as "the most measured, pro-AI mathematician" (12:40 AM ET Sep 9).

What's actually new: the congrats framing. The community is crediting Alpöge and Buckmaster with developing the AI-assisted technique — which keeps the attribution question open regardless of what the OpenAI agents did or didn't read. (Full treatment in today's ai digest.)

Links: - Malik RT: https://x.com/ylecun/status/2097709811894800397 - Kempe RT: https://x.com/ylecun/status/2097715395633295707

Perplexity × Nous Research: search as agent infrastructure (continuation, lead for the fresh item)

What happened: Arav amplified the Nous Research integration — "You can now use the Perplexity Search API in Hermes Agent" (8:40 PM ET Sep 8) — with the index claim: "Perplexity's index currently includes 450B+ high-quality URLs. Rapidly advancing to a trillion by EOY" (10:24 PM ET Sep 8). Then the freshest item in the window: the Q2D-Web RT — a benchmark and public leaderboard for evaluating retrieval in agentic RAG systems (4:21 PM ET).

What's actually new: the distribution shape. Perplexity is positioning its index as the retrieval substrate for third-party agents (Nous's Hermes is an open-weights agent), with a benchmark to set the eval terms. The "trillion by EOY" index claim is a dated, checkable promise — logged.

Sentiment check: Arav on Nous/Teknium — "relentless product execution and taste... glad to have Perplexity" (2:18 AM ET). Partnership warmth, not news.

Links: - Hermes integration: https://x.com/AravSrinivas/status/2097531833533771993 - Index claim: https://x.com/AravSrinivas/status/2097511314260414768 - Q2D-Web: https://x.com/AravSrinivas/status/2097799033452642378

Images 2.5's viral-benchmark streak: the scrambled Rubik's Cube mirror test (continuation)

What happened: steipete RT'd a viral claim (surfaced in feed Sep 9, posted 3:11 PM ET Sep 8): ChatGPT Images 2.5 "just passed the scrambled Rubik's Cube mirror test. The reflection is actually correct."

What's actually new: another round in Images 2.5's unofficial visual-benchmark tour — vendor-adjacent by construction, same genre as the Andon Labs RT claims we've been tracking in the Astra thread. Contrast remains the interesting bit: sama's own modest line that Images 2.5 "can't solve super difficult math problems" (Sep 8) while the model racks up spatial-reasoning party tricks. No independent harness behind the mirror test.

Links: - steipete RT: https://x.com/steipete/status/2097746170164715956

Coding-on-the-phone prediction (sentiment)

What happened: nichochar (Claude Code) quoting "JC": "a lot more coding is happening on the phone, but screen size does make the flow limited... A foldable iPad Mini, connected to an army of agents. The iPhone Duo will be the #1 coding platform in 2027" (4:06 PM ET).

What's actually new: the agent-era form-factor bet — the phone (or foldable) as remote-control for an agent fleet rather than the coding surface itself. It's a prediction by association (nichochar quoting someone else), so recorded as sentiment, not a ledger claim. Notable only because it's the Claude Code lead saying the IDE's successor isn't a screen he can ship.

Links: - https://x.com/nichochar/status/2097778653530898542

Quick hits

Dropped, and why