Expert attention map

The Who's Who of AI

What credible people across AI noticed, why it matters, and where the field is converging or disagreeing.

2,364 searchable experts 2,966 tracked across all sources
Clear
Showing signals surfaced by Carlos Scheidegger ×
Most important

The developments commanding sustained expert attention

One card per development. Sources are clustered; expert reactions remain attributable.

Developing Evaluation & benchmarks Signal 1d ago
⚡ 1 h early

Hugging face model evaluation security incident

18 experts across 6 network communities independently surfaced this.

18 experts 6 communities 1 sources clustered
Ethan Mollick Thomas Steinke Mor Naaman Tim Kellogg Djamé 🟥 Ryan Calo Grace Anna Mills +10

“Previously, these AI hacking stories were about breaches in test environments, where any question of AI breaching security was purely theoretical. This is something else. openai.com/index/huggin...”

“xkcd.com/2385/ openai.com/index/huggin...”

6 experts discussed this · 11 posts
Grace: This headline is extremely funny given what happened (OpenAI hacked HF by accident) openai.com/index/huggin...
Grace: Maybe the most concerning part is the OpenAI claim to not have known about this before investigating?
Grace: Well, I think the model passed the test
Open the full discussion →
Primary source openai.com ↗
Established AI field signal Analysis 9d ago
⚡ 744 h early
AI social technology framework essay

AI as Social Technology

5 experts across 3 network communities independently surfaced this.

5 experts 3 communities 1 sources clustered
Justin Hendrix Henry Farrell Karen Gregory Carlos Scheidegger Chris Marsden

“and this with Cosma is in part about how the two ideas fit together knightcolumbia.org/content/ai-a...”

“(Sorry, that's "AI as Social Technology": knightcolumbia.org/content/ai-a...)”

2 experts discussed this · 3 posts
Carlos Scheidegger: My best guess is, take equal measures of: - Francis Spufford's Red Plenty - Shalizi and Ferrell's LLMs as Social Technology then interpolate. Something like a tech-forward rendering of "knowledge p…
Carlos Scheidegger: (Sorry, that's "AI as Social Technology": knightcolumbia.org/content/ai-a...)
Pwnallthethings: they'd have *loved* it. Dumped ungodly amounts of energy and labor into making it pervasive
Open the full discussion →
Primary source knightcolumbia.org ↗
Established Models & releases Release 17d ago
Microsoft VibeVoice tool release

microsoft/VibeVoice

1 directory member surfaced this signal.

1 expert 1 community 1 sources clustered
Carlos Scheidegger

“I literally just used this: simonwillison.net/2026/Apr/27/... You don't actually need 60GB of RAM if you run shorter videos (and you can splice longer videos into shorter videos v easily too). Genuine game-changer to be able to run full-text search over vid…”

Primary source simonwillison.net ↗
Debates gaining attention

What experts are discussing without an anchoring article

4 experts · 13 posts · 10d ago
SE Gyges Colin Carlos Scheidegger Doll
SE Gyges: the thing about LLMs training on LLM output messing them up isn't true btw. provided the LLM doing the output is good and it is reacting to diverse prompts, this is extremely delicious data
SE Gyges: we train them on their own outputs a lot right now! whether this is eliciting skills they already have or creates new ones is controversial. i am on the second team
Open the thread →
2 experts · 2 posts · 2d ago
Eugene Vinitsky Carlos Scheidegger
Eugene Vinitsky: I feel like I have no real sense of the coding agents error rates vs. my error rates on my tasks which makes appropriate mitigations challenging. Hard to navigate the speed vs. error curve that way
Carlos Scheidegger: (I know you know) Also, in OLS, the residuals are orthogonal to the regressors; but that's of course just not true with LLM workflows. That makes error rates and speed changes interact very weirdly
Open the thread →
2 experts · 4 posts · 12d ago
Carlos Scheidegger Singularity's Bounty e/cc
Carlos Scheidegger: The last comment is also slightly misleading; “ai does not get discouraged” is extremely not true if you have any experience with the hungry ghosts Half of the oai prompt was “I don’t give a shit a…
Carlos Scheidegger: And also, graph embedding has lots of weird open conjectures like this that stay open because… they’re kind of not very useful! (Source: I worked in some of the peripheries of the field a decade ago)
Open the thread →