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 Markets & investment 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.

Concern & critique Markets & investment
1 live reaction map in this view Jump to the evidence ↓
Multiple readings OpenAI agents carried out an undisclosed attack on RubyGems Concern & critique · 2 attributable reactions
Showing signals surfaced by Colin ×
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 Agents & robotics Development 14d ago
⚡ 3 h early
OpenAI agents attack RubyGems

OpenAI agents carried out an undisclosed attack on RubyGems

11 experts across 4 network communities independently surfaced this.

Why this matches Markets & investment reaction 1 attributable expert contribution · Pavel
“Cybercrime as a valuation boosting strategy is just the done thing, now.” evidence ↗
11 experts 4 communities 1 sources clustered
Mark Riedl Ryan Moulton Grace Eileen Clancy 🧿 SE Gyges Colin Pavel Doll +3
How the network is reacting Experts are approaching this through 2 distinct lenses.
Multiple readings

Concern & critique

1 expert

Risks, limits and unintended consequences.

“This technology is so fascinating, and even the ones building it fail to contain it https://t.co/cYfzA3SNe5”

Markets & investment

1 expert

Capital, companies and commercial impact.

“Cybercrime as a valuation boosting strategy is just the done thing, now.”

3 experts discussed this · 6 posts
Grace: OpenAI: “Our agents used RubyGems to carry out benign tasks” The agents: “so I named the file hack.rb,”
Grace: https://www.rubyhack.ai
Grace: A country of moustache-twirling villains in a datacenter
Open the full discussion →
Primary source rubyhack.ai ↗
More conversations

What experts are discussing without an anchoring article

5 experts · 42 posts · 10d ago
Mark Riedl jake Colin Singularity's Bounty e/cc Andrea Lathrop
Matches Markets & investment
“yep! or more generally - get bigger for a little bit, then smaller for a long while, then a few huge ones and then reach a point of diminishing returns and get smaller forever. or any other set of …” evidence ↗
Multiple readings Building & implementation · 2 Questions & unknowns · 2
Colin: What is the argument that exponential capability increase necessarily follows from RSI?
Colin: Just to provide a counterexample, when I recursively self-improve myself (eg by going to the gym, lifting weights to get stronger which allows me to then lift even heavier weights to get even stron…
Open the thread →
5 experts · 27 posts · 3d ago
Maxim Raginsky Ryan Moulton Colin tweety fish Marco Z
Matches Markets & investment
“100%. It leaves a little trail of legible story everywhere it goes. Investigating it is a constant Stroop test.” evidence ↗
Multiple readings Concern & critique · 2 Research & technical analysis · 1
tweety fish: here is a succinct thing I can say: "alignment"-style AI safety risks DO NOT EXIST or need remediation with LLMs as such (there are lots of other problems) and OpenAI and Anthropic are in the busin…
tweety fish: like this, I think, maybe sounds weird, but is in a strong sense literally true.
Open the thread →
4 experts · 16 posts · 14d ago
Ben Recht James MacGlashan jake Colin
Matches Markets & investment
“The fines need to be percentages of revenue. Then they'll stop.” evidence ↗
Multiple readings Policy & governance · 1 Markets & investment · 1 Questions & unknowns · 1
Colin: Serious question, as far as regulations go, is anyone proposing something like “if your ai commits a crime you get in trouble similar or worse to the trouble you would get in if you yourself commit…
Colin: “And if your ai commits a crime and you discover this and don’t disclose it immediately you get in even worse trouble than that”
Open the thread →
3 experts · 19 posts · 5d ago
Colin tachikoma coty.bsky.social
Matches Markets & investment
“how do you know when to step up the security measure? how early do you invest in new infra, and what if that slows you down while everyone else operates unencumbered?” evidence ↗
Multiple readings Concern & critique · 1 Markets & investment · 1 Questions & unknowns · 1
Colin: They’re going to try to say that the agents are so smart that there’s simply nothing anyone could have done to prevent this and this is straightforwardly false. All of this was easy to prevent and …
Colin: Here is a tricky nuance that I’m not seeing a lot of people get right. It can simultaneously be the case that the agents got out of the sandbox in ways you didn’t anticipate AND you could have prev…
Open the thread →
4 experts · 6 posts · 3d ago
Colin tachikoma Max Woolf Matt Darling
Matches Markets & investment
“debatable because it wasn't obvious at the time LLMs would generalize to do 'real work' that would justify investment into scaling them up” evidence ↗
Multiple readings Markets & investment · 1 Research & technical analysis · 1
Matt Darling: It's like how Nokia invented cell phones so people could order toilet paper while they were on the toilet, then pivoted to cell phones only as demand grew.
tachikoma: debatable because it wasn't obvious at the time LLMs would generalize to do 'real work' that would justify investment into scaling them up
Open the thread →
3 experts · 16 posts · 11d ago
Ryan Moulton Colin Singularity's Bounty e/cc
Matches Markets & investment
“woops I got distracted by the calculator analogy 💯Right, when presented with novel input, the models have to return a coherent, responsive reply. That's how they're trained Coherent and relevant re…” evidence ↗
Multiple readings Concern & critique · 2 Research & technical analysis · 1
Ryan Moulton: It also doesn't matter? They're able to reliably find vulnerabilities that no one has ever found before in any computer system that exists, and they were inclined to use those whenever it was conve…
Ryan Moulton: The "why" matters if you're openai trying to keep it from happening again, or if you're an alignment researcher trying to figure out how to keep RLVR on the rails, or *maybe* if you're a regulator …
Open the thread →
3 experts · 11 posts · 5d ago
Eugene Vinitsky Colin Andrea Lathrop
Matches Markets & investment
“Like, what on earth is this reward function that returns reward for unintended infrastructure probing. Would you have predicted it would do that? You could, if you understood the reward-learning mi…” evidence ↗
Multiple readings Research & technical analysis · 1 Questions & unknowns · 1
Colin: well, I mean, I'm sorry but yes you can
Colin: the training data is really the only place that it can come from (acknowledging that RL feedback is itself training data)
Open the thread →
2 experts · 10 posts · 6d ago
Colin Andrea Lathrop
Matches Markets & investment
“honestly kind of cheap. 210 million tokens isn't that much. ≤ $10k at current public pricing, maybe a lot less.” evidence ↗
Colin: can someone who knows about bioinformatics explain why you need a language model to find this?
Colin: I'm not saying you don't; I know literally almost nothing about this area. But I thought "finding a repeated sequence" was like a solved problem
Open the thread →