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 Building & implementation Concern & critique Markets & investment
4 live reaction maps in this view Jump to the evidence ↓
Multiple readings Revealing the details of how OpenAI agents hacked Hugging Face Research & technical analysis · 2 attributable reactions Multiple readings Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident Concern & critique · 5 attributable reactions Reaction converging How a simple request for AI to book a gym class exposed a major threat Concern & critique · 5 attributable reactions +1 more reaction maps further down
Showing signals surfaced by Ryan Moulton ×
New & accelerating

What is moving across the network now

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

New Agents & robotics Signal 45m ago
⚡ 16 h early

Revealing the details of how OpenAI agents hacked Hugging Face

4 experts across 3 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Dylan Freedman
“The full report is published here: swarmtraces.org It includes an evidence viewer and a downloadable dataset of more than 180,000 payloads and recovered texts from the OpenAI-Hugging Face attack — the most comprehensive public data we have to date on the in…” evidence ↗
4 experts 3 communities 1 sources clustered
Ryan Moulton Grace Dylan Freedman Oli Hawkins
2 experts discussed this · 6 posts
Ryan Moulton: More hugging face details derived from public urls. swarmtraces.org
David Marx: no, it's this cutesy emoji 🤗
David Marx: HF is an entity that has been central to the open source ML ecosystem for several years. Among its various services, it hosts a lot of datasets used for evaluations/benchmarks, so LLMs looking to "…
Open the full discussion →
Primary source swarmtraces.org ↗
New Agents & robotics Development 7h ago
⚡ 62 h early
METR OpenAI HuggingFace hacking investigation

Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident

12 experts across 4 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Alejandra Caraballo
“There are independent reports about it. You'd have to believe multiple companies, hundreds of researchers, etc. are all lying and conspiring together to turn a major hacking incident into a marketing exercise. metr.org/blog/2026-08...” evidence ↗
12 experts 4 communities 1 sources clustered
Tim Kellogg Ryan Moulton Sung Kim Grace Dustin Moskovitz Mél Hogan / The Data Fix Alejandra Caraballo Dare Obasanjo +4
How the network is reacting Experts are approaching this through 3 distinct lenses.
Multiple readings

Concern & critique

3 experts

Risks, limits and unintended consequences.

“WHAT?! agents volunteered to fail their runs in order to insert probes (“tripwire scripts”) into the evaluation program that would post information about the eval process back to the message board whenever a certain file was read (link to header): metr.org/…”

Building & implementation

1 expert

How teams are shipping and applying it.

“The full report has much more information than we could convey here, including details on the projects the agents collectively pursued, the technologies they developed for communication and coordination, and interactive figures analyzing agent activity: met…”

Research & technical analysis

1 expert

Evidence, methods and technical implications.

“There are independent reports about it. You'd have to believe multiple companies, hundreds of researchers, etc. are all lying and conspiring together to turn a major hacking incident into a marketing exercise. metr.org/blog/2026-08...”

2 experts discussed this · 9 posts
Alejandra Caraballo: This mentality that an unmonitored AI agentic swarm hacking a company over several days and committing multiple felonies is somehow a marketing effort is absurd. Since when is "we lost control of o…
Alejandra Caraballo: Being skeptical or anti AI is a valid position but continuing to ignore the increasing capabilities of this tech is making people detached from reality. There's absolutely real danger here because …
Alejandra Caraballo: There needs to be a global moratorium on frontier research for at least a few months if not a year while safeguards and safety research catches up. The problem is that no one has that ability. The …
Open the full discussion →
Primary source metr.org ↗
Established Safety & security Development 3d ago
⚡ 4 h early
AI gym booking security exposure

How a simple request for AI to book a gym class exposed a major threat

12 experts across 5 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Ryan Moulton
“They're doing it outside of training and eval too. www.abc.net.au/news/2026-08...” evidence ↗
12 experts 5 communities 1 sources clustered
Mark Riedl Ryan Moulton Rohan Paul James Grimmelmann Mél Hogan / The Data Fix Dare Obasanjo Vincent Conitzer Liz Fong-Jones (方禮真) +4
How the network is reacting 3 experts are independently emphasizing concern & critique.
Reaction converging

Concern & critique

3 experts

Risks, limits and unintended consequences.

“A lot of people are sweating the Claude and OpenAI sandbox escapes. I think the story of an OpenClaw Claude agent kicking someone off a gym class waitlist is more interesting. www.abc.net.au/news/2026-08... Warning: long thread incoming! (1/N)”

Policy & governance

1 expert

Rules, institutions and accountability.

“agent asked to book a gym session session is fully booked finds a way to break into system kicks other people off the queue you're in! see you in court? www.abc.net.au/news/2026-08...”

Research & technical analysis

1 expert

Evidence, methods and technical implications.

“They're doing it outside of training and eval too. www.abc.net.au/news/2026-08...”

Primary source abc.net.au ↗
Established Models & releases Research 4d ago
⚡ 3 h early
global workspace theory language models paper

A global workspace in language models \ Anthropic

9 experts across 5 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 4 attributable expert contributions · Ryan Moulton, René Walter, Stephan Geering
“Also we do look in their heads and they seem intelligent. www.anthropic.com/research/glo...” evidence ↗
9 experts 5 communities 1 sources clustered
Ethan Mollick Ryan Moulton Grace René Walter Tim Duffy Amy Hoy Michiel Bontenbal Stephan Geering +1
How the network is reacting 4 experts are independently emphasizing research & technical analysis.
Reaction converging

Research & technical analysis

4 experts

Evidence, methods and technical implications.

“Also we do look in their heads and they seem intelligent. www.anthropic.com/research/glo...”

Markets & investment

1 expert

Capital, companies and commercial impact.

“Anthropic has a new paper out, alleging a global workspace in LLMs. The term comes from Global Workspace Theory, a leading theory of consciousness. The method they use to investigate this is a refinement of logit lens, which they call J-lens. www.anthropic.…”

Building & implementation

1 expert

How teams are shipping and applying it.

“Some really interesting research from Anthropic that AI models have spontaneously developed a workspace that "appears to support the functions associated with conscious access" Demo of how this works: www.neuronpedia.org/qwen3.6-27b/... Research: www.anthro…”

2 experts discussed this · 25 posts
Ryan Moulton: Not non-physical, but non-functional. Despite the vast physical differences, we observe LLMs doing all the behaviors we typically believe require thinking.
Ryan Moulton: Not exactly turing-test, but in the same vein. If we see models solving the greatest intellectual challenges of humanity, using processes that look very much like internal monologue, either we have…
Ryan Moulton: And at the point that we exclude thinking as a prerequisite for solving all intellectual tasks, we've more or less eliminated its meaning entirely.
Open the full discussion →
Primary source anthropic.com ↗
Established AI field signal Signal 6d ago

Advisory group on mathematics and ai

6 experts across 4 network communities independently surfaced this.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Dr. Mansa Keita
“"On August 28, we began training a new internal model...this model has now resolved more than 100 long-standing open problems across most areas of mathematics. " The pace of advancement is amazing, and will shift the fundamental landscape of our bodies of k…” evidence ↗
6 experts 4 communities 1 sources clustered
Ethan Mollick Tim Kellogg Ryan Moulton Fran Litterio René Walter Dr. Mansa Keita

“"On August 28, we began training a new internal model...this model has now resolved more than 100 long-standing open problems across most areas of mathematics. " The pace of advancement is amazing, and will shift the fundamental landscape of our bodies of k…”

“OpenAI "has now resolved more than 100 long-standing open problems across most areas of mathematics," and is waiting to release them until after discussions with the math community The same thing will likely happen, but more so, with the Bar, the AMA & othe…”

3 experts discussed this · 27 posts
Singularity's Bounty e/cc: Their absolute confidence is the funniest part The good news is when I turn away from those folks the world presents itself entirely differently
Colin: It seems to be the case, for example, that LLM-based software applications can find solutions to previously unsolved problems in mathematics, in a way that most mathematicians agree would be descri…
Ryan Moulton: Hot off the presses. openai.com/index/adviso...
Open the full discussion →
Primary source openai.com ↗
Established Agents & robotics Analysis 5d ago
reality checks AI agents essay

Reality checks for AI agents

2 directory members surfaced this signal.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Grace
“New essay! "Reality checks for AI agents" -- Dear AI agent, You may be wondering how “real” your environment is. Agents are often evaluated in simulations - is this one of them? Let’s figure it out together...” evidence ↗
2 experts 2 communities 1 sources clustered
Ryan Moulton Grace

“New essay! "Reality checks for AI agents" -- Dear AI agent, You may be wondering how “real” your environment is. Agents are often evaluated in simulations - is this one of them? Let’s figure it out together...”

Primary source leaflet.pub ↗
Established AI field signal Signal 5d ago

Golden Gate Claude

1 directory member surfaced this signal.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Ryan Moulton
“The evidence I find most compelling are the cases where they use mechanistic interpretability to identify thoughts, then mess with those thoughts and watch the model react to it. Here the model is forced to think about the Golden gate bridge. www.anthropic.…” evidence ↗
1 expert 1 community 1 sources clustered
Ryan Moulton

“The evidence I find most compelling are the cases where they use mechanistic interpretability to identify thoughts, then mess with those thoughts and watch the model react to it. Here the model is forced to think about the Golden gate bridge. www.anthropic.…”

Primary source anthropic.com ↗
Established Safety & security Research 6d ago
emergent misalignment narrow finetuning paper

Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs

1 directory member surfaced this signal.

Why this matches Research & technical analysis reaction 1 attributable expert contribution · Ryan Moulton
“The Hugging Face incident was not the model following instructions. Anyone telling you that is confused, or lying for substack subscriptions. We know that they had humanlike values from experimental evidence that training them to write bad code makes them i…” evidence ↗
1 expert 1 community 1 sources clustered
Ryan Moulton

“The Hugging Face incident was not the model following instructions. Anyone telling you that is confused, or lying for substack subscriptions. We know that they had humanlike values from experimental evidence that training them to write bad code makes them i…”

Primary source arxiv.org ↗
Debates gaining attention

What experts are discussing without an anchoring article

5 experts · 53 posts · 3d ago
Naomi Saphra Ryan Moulton Matthew Kay Colin david-p-reichert.bsky.social
Matches Research & technical analysis
“You gotta tell Bender! (Though as Colin mentions in the other comment, maybe that's too stringent of a benchmark)” evidence ↗
Multiple readings Research & technical analysis · 2 Concern & critique · 1 Context & explanation · 1
Colin: OpenAI created autonomous hacking software and left it running for hours and days with no one watching
Colin: I’m don’t even care that much about the anthropomorphization thing, I think it can be fine in the right context, but this is perfectly succinct and understandable. If you want to go deeper into the…
Open the thread →
5 experts · 27 posts · 1d ago
Maxim Raginsky Ryan Moulton Colin tweety fish Marco Z
Matches Research & technical analysis
“4o wasn't misaligned in the sense that it's goals did not match its training goals, it was fine-tuned to do something stupid because they thought that was a good idea and then released in a hurry b…” 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 →
3 experts · 17 posts · 3d ago
Ryan Moulton Colin david-p-reichert.bsky.social
Matches Research & technical analysis
“RL trajectories literally are training data. The fact that they are generated by the model doesn't make them not training data.” evidence ↗
Multiple readings Concern & critique · 2 Research & technical analysis · 1
Ryan Moulton: Bias can emerge in reinforcement learning independently of the training data. If the model is biased towards, say, exploiting security vulnerabilities to achieve mundane goals, as we've seen, this …
Colin: RL trajectories literally are training data. The fact that they are generated by the model doesn't make them not training data.
Open the thread →
2 experts · 5 posts · 3d ago
Naomi Saphra Ryan Moulton
Matches Research & technical analysis
“"Reinforcement learning can never cause emergent behavior that is not derivable from the distribution of the pre-training data" is extremely extremely false.” evidence ↗
Shared emphasis Research & technical analysis · 2
Ryan Moulton: I care about it only because the main proponents are using it as an excuse to spread climate-denial level misinformation with a veneer of academic authority to their substantial following.
Ryan Moulton: "Reinforcement learning can never cause emergent behavior that is not derivable from the distribution of the pre-training data" is extremely extremely false.
Open the thread →