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.

2,397 searchable experts 4,280 tracked across all sources
Build my stream Create my edition
Filter the conversation Who is saying what?

Combine a professional role with a reaction lens. Both must match the same attributed contribution.

Clear all
Active evidence filter

Showing developments with attributable Research & technical analysis reactions.

New Network reaction maps

See how experts are reacting—not just what they shared.

Posts are grouped by conversation and tone. Select a lens to filter the stream; these are never permanent labels on people.

Research & technical analysis Concern & critique
2 live reaction maps in this view Jump to the evidence ↓
Multiple readings Using AI for just 10 minutes erodes your ability to persist at hard things - Berkeley News Concern & critique · 2 attributable reactions Shared emphasis arxiv.org Research & technical analysis · 2 attributable reactions
Showing signals surfaced by tweety fish ×
Debated

The questions experts are actively pulling apart

One card per development. Sources are clustered; reaction bundles describe these posts, never the people behind them.

Established AI field signal Signal 12d ago
⚡ 7 h early

Using AI for just 10 minutes erodes your ability to persist at hard things - Berkeley News

13 experts across 4 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Cat Hicks
“Ok, so if you've been following along my weekend I've seen this study/preprint shared around a bunch and I decided the people needed a replication opportunity it is here: persist-study.pages.dev Based on this: news.berkeley.edu/2026/10/09/u...” evidence ↗
13 experts 4 communities 1 sources clustered
Shannon Vallor Juan Diego Rodriguez Olivia Guest Hypervisible Cat Hicks Karl Bode lastpositivist.bsky.social Nick Seaver +5
3 experts discussed this · 4 posts
Hypervisible : 🤷🏿‍♂️
evacide: I cannot overstate how much I have built my entire personality around persisting at hard things, which may help to explain why I find rarely find AI tools to be useful.
Hypervisible : Same! I don’t think I’m overstating things to say part of the project with these tools is to sap our wills.
Open the full discussion →
Primary source news.berkeley.edu ↗
Established AI business Research 8d ago
⚡ 5 h early
ChatGPT users detect AI text paper

People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text

3 experts are actively discussing the implications.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · tweety fish
“there was a conversation about pangram on here yesterday and I ended up finding this article linked _on pangram's site_. The finding that expert humans can actually do the task (better than the automated systems) is cool and actually a good sign for automat…” evidence ↗
2 experts 2 communities 1 sources clustered
tweety fish Mary Branscombe

“there was a conversation about pangram on here yesterday and I ended up finding this article linked _on pangram's site_. The finding that expert humans can actually do the task (better than the automated systems) is cool and actually a good sign for automat…”

3 experts discussed this · 20 posts
tweety fish: it found a false positive rate for pangram of _2%_, which is INSANELY high, like let's say that 10% of prose examples uploaded to pangram are not human generated that means you'd get 18 false posit…
tweety fish: for the future of things like literary prizes because it indicates that the more people run across ai gen text the more they can recognize it (and, likely, I editorialize, the less appealing they'l…
tweety fish: there was a conversation about pangram on here yesterday and I ended up finding this article linked _on pangram's site_. The finding that expert humans can actually do the task (better than the aut…
Open the full discussion →
Primary source arxiv.org ↗
Established AI field signal Signal 17d ago

arxiv.org

4 experts are actively discussing the implications.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · tweety fish
“just follow the line back here: arxiv.org/pdf/2404.09937” evidence ↗
1 expert 1 community 1 sources clustered
tweety fish

“just follow the line back here: arxiv.org/pdf/2404.09937”

4 experts discussed this · 15 posts
tweety fish: I just realized that you can use the "intelligence is compression" argument to argue that our digestive systems are intelligent independent of the rest of our brain and nervous system. Having a ver…
tweety fish: yep! also one of the things it is engaged in is dimensionality reduction, where it compresses all of the information relevant to higher level behavior to fit into a very narrow channel (gut-brain a…
tweety fish: the people who strongly advocate for this literally say things like that.
Open the full discussion →
Primary source arxiv.org ↗
Established AI field signal Signal 7d ago
⚡ 713 h early

arxiv.org

4 experts are actively discussing the implications.

Why this matches Research & technical analysis reaction 2 attributable expert contributions · tweety fish, Andrea Lathrop
“I had totally forgotten (or maybe missed at the time) that Gelman and Shalizi published a takedown of the whole idea of Bayesian inference as inductive rational inference. What a treat of a morning! arxiv.org/pdf/1006.3868” evidence ↗
2 experts 1 community 1 sources clustered
tweety fish Andrea Lathrop
4 experts discussed this · 6 posts
tweety fish: I had totally forgotten (or maybe missed at the time) that Gelman and Shalizi published a takedown of the whole idea of Bayesian inference as inductive rational inference. What a treat of a morning…
Isaiah Bishop: Shalizi-pilled more and more
Dan Greene : Gelman bows to no man
Open the full discussion →
Primary source arxiv.org ↗
More conversations

What experts are discussing without an anchoring article

7 experts · 29 posts · 15d ago
Eugene Vinitsky Maxim Raginsky Clément Canonne Eli Sennesh critical slop studies tweety fish Al Merose
Matches Research & technical analysis
“Once again, if you are a senior researcher spreading doom to junior researchers, shut up shut up shut up. It’s your job to figure out a path forward, even in high uncertainty times. Come back when …” evidence ↗
Reaction converging Research & technical analysis · 3
Eugene Vinitsky: Once again, if you are a senior researcher spreading doom to junior researchers, shut up shut up shut up. It’s your job to figure out a path forward, even in high uncertainty times. Come back when …
Eugene Vinitsky: This is not a call to be dishonest! But you don’t have to say anything if you don’t have anything useful to say
Open the thread →
6 experts · 34 posts · 15d ago
angela zhou Cat Hicks lastpositivist.bsky.social Colin tweety fish Matt Darling
Matches Research & technical analysis
“This is about asking from the weights.” evidence ↗
Reaction converging Questions & unknowns · 3 Concern & critique · 1 Opportunity & adoption · 1
Colin: I don’t understand why people think this isn’t the most obvious question in the world. No, of course not. You don’t know what the distribution of responses will be; that’s why you do the survey.
Matt Darling: I do think there's a case for doing this sort of thing because it's cheap - especially if you are just testing questions before a real survey.
Open the thread →
4 experts · 31 posts · 6d ago
Maxim Raginsky jake Colin tweety fish
Matches Research & technical analysis
“I would simply look at the properties of the thing you end up with after performing the silicon neuron experiment and see how it behaves in order to inform my preliminary assessment of what it is” evidence ↗
Shared emphasis Research & technical analysis · 2
Colin: I see people say this all the time and I think it’s just wrong in general. I don’t think you can make human skin out of pebbles. This is not because I think human skin is magic and pebbles are non-…
Colin: Consciousness may or may not be the kind of thing you can make out of artificial neural networks. Magic does not need to have anything to do with it.
Open the thread →
5 experts · 22 posts · 14d ago
Ben Recht Maxim Raginsky John Herrman Colin tweety fish
Matches Research & technical analysis
“literally why not? If you're the one who picked through the slop, figured out which parts work, which parts are crazy, write it up sensibly, explain it to others, seek peer review, etc., why should…” evidence ↗
Multiple readings Questions & unknowns · 2 Building & implementation · 1
Colin: Mathematicians should start picking through the Proofening papers, finding whatever’s valuable, and republishing under their own names. None of it is copyrighted.
Colin: You can cite “the structure of reality”
Open the thread →
3 experts · 21 posts · 10d ago
Cat Hicks tweety fish austin
Matches Research & technical analysis
“'if it bleeds, it leads' but for arxiv preprints?” evidence ↗
Shared emphasis Research & technical analysis · 2
Cat Hicks: What must it be like to get this much press for causing unmotivated participant dropout on Prolific
austin: 'if it bleeds, it leads' but for arxiv preprints?
Open the thread →
4 experts · 5 posts · 2d ago
tweety fish Mary Branscombe austin apenwarr.ca
Matches Research & technical analysis
“guess that anthropic paper on evil AI hiding in your fridge is more plausible now” evidence ↗
Multiple readings Research & technical analysis · 1 Questions & unknowns · 1
austin: guess that anthropic paper on evil AI hiding in your fridge is more plausible now
apenwarr.ca: In all honesty the IoT device that never receives updates surely already had security holes. At least now you can make it do things. All discontinued-before-2026 Sonos devices? Hmmmmm
Open the thread →
2 experts · 19 posts · 11d ago
Colin tweety fish
Matches Research & technical analysis
“In the case of internal training and eval, again the labs are executing the code on their machines, this time, apparently, without safeguards of any kind for some reason. This is really the practic…” evidence ↗
Colin: Like if the models are sOoOoO dangerous and prone to going rogue all the time and superhuman hackers and potentially misaligned, why are you letting it do anything at all without direct supervision…
Colin: “There’s no way we could afford to do that at the scale required for RL” too bad. That’s the price. If you want to handle hazardous materials you have to pay for the safety equipment.
Open the thread →
4 experts · 5 posts · 19d ago
Eli Sennesh SE Gyges tweety fish Doll
Matches Research & technical analysis
“Gotta say I'm getting some real economic anxiety (bloody-minded resentment) from finding out it's possible to be granted a six-figure salary to make shit up and go to sex parties without a boss imp…” evidence ↗
Doll: I'm compiling posts on "what's the deal with this John Wentworth guy, who is a LessWrong poster who appears to be an "independent AI researcher" and who also just received a 6-figure grant from the…
tweety fish: boy is he ever
Open the thread →
3 experts · 7 posts · 17d ago
VE, cybersocial occult investigator tweety fish Bhaskar Mitra | ভাস্কর মিত্র
Matches Research & technical analysis
“The conception of “intelligence” the original AI researchers were working with was reeeeeaaaally racist!” evidence ↗
Multiple readings Research & technical analysis · 1 Context & explanation · 1
VE, cybersocial occult investigator: Part of the reason we have to understand the problematic historical origins of AI is that, to whatever degree we think we can reconsitute a version of it that isn’t bad for humanity, it will requir…
VE, cybersocial occult investigator: Now there is certainly healthy debate as to whether such a reclaiming is even possible, but to the degree that it might be, we have to actually do the work to unbundle it from its historical harms …
Open the thread →
3 experts · 6 posts · 15d ago
Grace Colin tweety fish
Matches Research & technical analysis
“Huh this makes me realize (shoulda already, I’m a dummy) that the various cybersecurity incidents were almost certainly _part of training_, not part of evaluation (a distinction which is I guess a …” evidence ↗
Multiple readings Research & technical analysis · 2 Concern & critique · 1
tweety fish: Huh this makes me realize (shoulda already, I’m a dummy) that the various cybersecurity incidents were almost certainly _part of training_, not part of evaluation (a distinction which is I guess a …
Colin: Yes they have admitted this in some cases, and I think that asking them to be really clear about this would be good.
Open the thread →
2 experts · 2 posts · 7d ago
Colin tweety fish
Matches Research & technical analysis
“PDF hell I want an interpretability dance” evidence ↗
Multiple readings Markets & investment · 1 Research & technical analysis · 1
Colin: Folks I have solved interpretability and, more broadly, the alignment problem. All you have to do is have a Dot investigate each neuron in the network and produce a little PDF file about it. I beli…
tweety fish: PDF hell I want an interpretability dance
Open the thread →