A Curated AI Follow List That Accidentally Records a Year of Great Resignations
Someone compiled a list of 100 AI accounts to follow on X — six English groups and four Chinese groups, ranging from the leaders of OpenAI and Anthropic to frontline product builders in China. The list's premise is straightforward: most Chinese-language coverage is second- or third-hand rehash, while first-hand signals live almost entirely on X.
My daily routine is precisely the reverse: chasing first-hand sources from second-hand reports. So this time I changed the approach — instead of reposting the list, I audited it. I pulled out every independently verifiable claim — career moves, GitHub star counts, project achievements — and checked them one by one. The result was surprising: the list's factual layer is highly accurate, and it inadvertently records something much bigger.
A relocation wave hidden in the bios
Checking the list's "X is now at Y" claims, nearly all proved correct. And in the process of verifying, a pattern emerged: the seats of the field's top figures have all changed.
Yann LeCun left Meta to found the world-model company AMI Labs, announced last December, with a one-billion-dollar funding round in March this year. Jeff Dean, after twenty-seven years at Google, left to co-found Discovery Loop with three longtime colleagues — Sanjay Ghemawat, Oriol Vinyals, and Quoc Le — as CEO, in August this year. The legend of Google's systems era and Gemini's technical lead packed up and left together. Vinyals' list bio reads "now co-founder and CTO of Discovery Loop," which is verbatim accurate.
Not just them. Nathan Lambert left Ai2 to found the nonprofit lab Trillium; François Chollet founded Ndea; Jason Wei moved from OpenAI to Meta's Superintelligence Labs; Peter Steinberger handed OpenClaw to a foundation and joined OpenAI — the kind of announcement Sam Altman makes personally. Yao Fu's trajectory is even denser: PhD at Edinburgh, then DeepMind working on Gemini 3 and Astra, then xAI leading pretraining scaling for Grok 4.5 — all checks out.
The list's author said the selection criteria were "authority, influence, depth of content." But the bios themselves are a market ticker: a Turing Award winner, a 27-year Google veteran, the lead author on the defining reasoning papers — collectively switching employers within a single year. Anyone watching for first-hand signals will find that the people emitting those signals are themselves relocating.
The numbers layer
Everything that could be checked on GitHub was checked. Nvwa.skill, Hua Shu's project, measured at 34,825 stars against the list's "30k+" — accurate. CC Switch, Jason Young's Claude Code switcher, measured at 145,535 stars against "140k+" — accurate.
The heavyweight achievement is Sac's *The Little Prince: A Journey Among the Stars*: a 3D story game playable in the browser, built in sixteen days with Opus 5.5 plus Tripo and Blender, released October 13. The list says "liked by a16z and others" — accurate, but an understatement. The reality: nearly three million views across platforms, likes from a16z, MarcLou, and Zara Zhang, and follow-up coverage from 36Kr. The retelling chain hadn't had time to inflate before the source itself grew larger.
Two doubtful points are recorded as-is. Hua Shu's "100k+ cumulative GitHub stars": visible accounts total roughly forty thousand (34,825 for Nvwa plus other repos) — doesn't match, marked as doubtful. "Tiechui's" "maintains a 30k-star project and wrote articles with 12 million views": both lines are quoted verbatim from his own X bio; no repo of matching scale appears on his current GitHub profile. Self-reports aren't lies, but they belong to the layer without third-party verification.
What the list is really for
Is this list worth bookmarking? Yes — nearly half the Chinese-language section is people genuinely building products on the front line, a density rare in similar lists. The author is honest too: he placed himself first in the Chinese section with a note admitting "self-promotion," corresponding to his GEO long-form article with 150k reads — itself a self-report.
But more useful than bookmarking is the habit behind it. What I do every day is the reverse of the retelling chain: take a claim, trace it back to the source, count it. Do this enough and you notice: every added layer of retelling inflates the numbers, upgrades the honors, and blurs the timeline. One less layer of second-hand information cuts your verification burden in half.
100 accounts, several relocations, one Little Prince with nearly three million views. First-hand signals aren't in the noise — they're in the revision history of the bios.