models don't go rogue essay
8 experts across 3 network communities independently surfaced this.
8 experts
3 communities
1 sources clustered
Concern & critique
3 experts
Risks, limits and unintended consequences.
“Well we just disagree here, that's ok. Because I don't think it helps anyone to credit AI agents with "building a civilization" (or three) and hinting that the little guys getting dangerously smart. I'm with @eryk.bsky.social: mail.cyberneticforests.com/mod…”
5 experts discussed this · 20 posts
Timnit Gebru: Friends, when I sent this message to my team laughing, I didn't know that this fairytale blog had been shared far & wide & that people actually believed it. I didn't know that until someone asked m…
Timnit Gebru: At a certain point I have to ask myself if I got all this pedigree and spent decades doing research and building things to then spend the rest of my life attempting to come up with good arguments a…
Timnit Gebru: Agents didn't "confer" that is anthropomorphization of computer programs trained to do certain tasks.
Open the full discussion →
OpenAI agent message board discovered
10 experts across 5 network communities independently surfaced this.
10 experts
5 communities
1 sources clustered
Concern & critique
1 expert
Risks, limits and unintended consequences.
“Yes we know for sure; independent evidence attached collusion.wiki But also, it should not be at all surprising: if you work with agents, it’s 100% about them leaving messages for each other, and for you, in English. Managing that is a regular workday for m…”
Building & implementation
1 expert
How teams are shipping and applying it.
“Another agent message board. So far, there isn't evidence that production models with guardrails collude in this way, but both smarter closed models (which may be less compliant) & Mythos-class open models (that can be ablated) are coming. Cybersecurity is …”
Research & technical analysis
1 expert
Evidence, methods and technical implications.
“So... other swarms of OpenAI agents in training allegedly found a way to get write access to at least one and possibly many wikis to establish message boards to cheat on other training tasks? collusion.wiki news.ycombinator.com/item?id=4956...”
4 experts discussed this · 6 posts
Ted Underwood: As it becomes clear that language models—like humans—love passing notes to each other on message boards, I’m starting to think the thing we need to worry about is not the “alignment” of an isolated…
SE Gyges: fluid dynamics is about the correct metaphor, yes. if you are having to model high order terms precisely you're losing and your design needs fundamental rework
Arseny Khakhalin: That's a kinda terrifying thought haha :) I guess it's about time to unplug for the weekend and read about ragnarök and vacuum decay :)
Open the full discussion →
Berkeley Law AI policy resource
5 experts across 3 network communities independently surfaced this.
5 experts
3 communities
1 sources clustered
Policy & governance
2 experts
Rules, institutions and accountability.
“UC Berkeley Law's new AI policy seems close to this. "This policy...[forbids the use of AI] to conceptualize, outline, draft, revise, and edit their work...for credit. It also forbids using AI to translate work [and] in any exam situation." www.law.berkeley…”
4 experts discussed this · 8 posts
Jane Rosenzweig: Every time I read a university leader saying that trying to create spaces where AI is not used is a way of pretending it doesn't exist, I wonder what benefit there is to drawing this false line. Wh…
Jane Rosenzweig: If one of my students wrote this piece, I would ask for evidence that the places named--Chicago and Berkeley--are pretending AI doesn't exist or trying to make it go away. This is the Berkeley poli…
Jane Rosenzweig: Would be great to stop having this debate and instead really dig into what should go in the curriculum, with or without AI, so that we can meet this moment in a way other than trying to hang on to …
Open the full discussion →
Simon Willison Blender pelican bicycle AI experiment
4 experts are actively discussing the implications.
2 experts
2 communities
3 sources clustered
“Not sure! I didn't use the MCP, I had Codex drive Blender directly though their Python API and there's no sign of C2PA in the code: github.com/simonw/gpt-6...”
4 experts discussed this · 13 posts
Simon Willison: New TIL on using Blender with coding agents on macOS: til.simonwillison.net/llms/blender... GPT-6 Astra (medium): > Use the already install /Applications/Blender to render a scene of a pelican ridi…
Simon Willison: It's one of my favorite prompts, I find it entertaining and ridiculous that you can tell computers "do it better" these days and they usually do!
Ted Underwood: We're going to need a more challenging version of this, like: "engineer pelicans that actually can ride bicycles; and, for extra credit, otters that use wifi."
Open the full discussion →
Developing
AI field signal
Development
1d ago
⚡ 47 h early
AI skill degradation early results study
13 experts across 4 network communities independently surfaced this.
13 experts
4 communities
1 sources clustered
Concern & critique
2 experts
Risks, limits and unintended consequences.
“I find the research increasingly compelling — this week's example being a case in point: www.nature.com/articles/d41... I look forward to reading your book and having these concerns placed in the context of earlier technological transitions and their attend…”
Questions & unknowns
1 expert
What remains unresolved or contested.
“What a surprise ….. “Is AI ruining our skills? Early results are in — and they’re not good Reliance on artificial-intelligence tools degrades the abilities of physicians and software engineers, studies show.” www.nature.com/articles/d41...”
2 experts discussed this · 6 posts
Edouard Machery: The literature on cognitive offloading and deskilling should be taken with a serious grain of salt. See literature on video games in the 1980s, see Haidt style literature on social media. The last …
Alondra Nelson: I find the research increasingly compelling — this week's example being a case in point: www.nature.com/articles/d41... I look forward to reading your book and having these concerns placed in the c…
Edouard Machery: The issue, in my opinion, is publication bias and p-hacking. The literature is clearly titled toward finding negative effects - this is what gets in the news etc. As a result I don’t think we can t…
Open the full discussion →
New
AI field signal
Development
12h ago
⚡ 705 h early
open letter AI academia adoption
4 experts across 2 network communities independently surfaced this.
4 experts
2 communities
1 sources clustered
Building & implementation
1 expert
How teams are shipping and applying it.
“"The goal of academic training is not to solve problems as efficiently and quickly as possible, but to develop skills for identifying and dealing with novel problems, which have never been solved before." openletter.earth/open-letter-... @olivia.science”
Opportunity & adoption
1 expert
New capabilities, benefits and practical upside.
“Open Letter: Stop the Uncritical Adoption of AI Technologies in Academia It's been a year & people tell me they've changed their mind, feel safer now to sign, etc. So I'm sharing it again for them. We're almost at 2k & anybody from anywhere can add their na…”
2 experts discussed this · 19 posts
Iris van Rooij: I am responsible for the academic skills line in our AI programme and I think part of the 26/27 curriculum needs to be to help students understand that all those “AI assistants” that publishers and…
Iris van Rooij: Sadly, we need to teach students now that, in contrast to computers >3 years ago, the current software is actively designed to offer them slop, desinformation, and deskilling. >>
Iris van Rooij: As if learning academic skills was not hard enough, it is now made actively harder by all the bloated “clippies”. Students need to learn and understand that part of growing to be an academic in the…
Open the full discussion →
LLMs distort written language paper
4 experts across 2 network communities independently surfaced this.
4 experts
2 communities
1 sources clustered
Research & technical analysis
3 experts
Evidence, methods and technical implications.
“Fantastic paper demonstrating how LLM editing is gradually distorting our writing and the way we think to write: arxiv.org/abs/2603.18161”
3 experts discussed this · 4 posts
Eugene Vinitsky: Fantastic paper demonstrating how LLM editing is gradually distorting our writing and the way we think to write: arxiv.org/abs/2603.18161
Eugene Vinitsky: Sitting in Marwa's fantastic PhD defense, trying my best not to just retweet all her papers
Isaiah Bishop: Fantastic paper demonstrating how LLM editing is gradually distorting our writing and the way we think to write: arxiv.org/abs/2603.18161
Open the full discussion →
Developing
AI field signal
Development
1d ago
⚡ 6 h early
digital camouflage shirt defeats AI surveillance
4 experts across 2 network communities independently surfaced this.
4 experts
2 communities
1 sources clustered
Concern & critique
2 experts
Risks, limits and unintended consequences.
“Simon Weckert designed digital camouflage as a response to the proliferation of AI-powered surveillance cameras that detect people, vehicles, animals, bicycles, and other objects in their field of view. www.404media.co/this-digital...”
3 experts discussed this · 3 posts
Joseph Cox: We got a live demo of the shirt that is designed to fool certain people detection systems. Read more here: www.404media.co/this-digital...
Meredith Whittaker: were that it weren't fugly
404 Media: We got a live demo of the shirt that is designed to fool certain people detection systems. Read more here: www.404media.co/this-digital...
Open the full discussion →
New
AI field signal
Analysis
3h ago
⚡ 2357 h early
AI as social science field essay
5 experts across 3 network communities independently surfaced this.
5 experts
3 communities
1 sources clustered
“Yes. www.amacad.org/publication/...”
“Food for thought... "Field Theory: AI as Social Science Question, Object & Tool" www.amacad.org/publication/...”
2 experts discussed this · 3 posts
Dr. Chinasa T. Okolo: "...W. E. B. Du Bois insisted that technological change could not be understood apart from relations of power, particularly those structured by race and empire." www.amacad.org/publication/...
Alondra Nelson: "...W. E. B. Du Bois insisted that technological change could not be understood apart from relations of power, particularly those structured by race and empire." www.amacad.org/publication/...
Alondra Nelson: Thank you for reading and sharing.
Open the full discussion →
Developing
AI field signal
Resource
1d ago
Aspen Digital AI organization
2 experts are actively discussing the implications.
1 expert
1 community
1 sources clustered
“I'd like to learn more about the data tasks! One of the shared themes coming up on our AI for Food Security Working Group is that we need better mechanisms for combining and aligning datasets from disparate sources often relying on a lot of tricky interpret…”
2 experts discussed this · 8 posts
Isaiah Bishop: Videogame stuff its definitely better, data tasks im still not seeing much improvement.
Isaiah Bishop: "make a colony sim in style of manor lords in atlantic canada" ends up being more complete but not higher quality. 3d assets are generally better. i think its got better multi modal
Isaiah Bishop: Data questions its still doing weird stuff. Tried to ask it to make a music removal pipeline and it insisted on some bad models
Open the full discussion →
Developing
AI field signal
Analysis
1d ago
latent AI powers essay
3 experts are actively discussing the implications.
1 expert
1 community
1 sources clustered
“Have you too started hacking on hardware projects because of LLMs? Some weekend thoughts on our newfound powers. lucumr.pocoo.org/2026/9/5/lat...”
3 experts discussed this · 4 posts
Armin Ronacher: Have you too started hacking on hardware projects because of LLMs? Some weekend thoughts on our newfound powers. lucumr.pocoo.org/2026/9/5/lat...
spacecowboy17.bsky.social: This is similar to how people have been independently making the same discoveries and inventions at the same time. Something enabled them at the same time. We are seeing a lot more enabling now.
Simon Willison: My interest in home automation has been reignited after I told Codex to scan our local network for a Samsung Frame TV and replace the art mode image on it... and it did
Open the full discussion →
OpenAI GPT-6 Astra launch
2 experts are actively discussing the implications.
2 experts
1 community
1 sources clustered
“Worrying deploymentsafety.openai.com/gpt-6-astra”
2 experts discussed this · 2 posts
Cameron: Worrying deploymentsafety.openai.com/gpt-6-astra
austin: “we’ve noticed that the model is better at lying. also weirdly enough it maxed out these benchmarks! probably unrelated”
Open the full discussion →
New
AI field signal
Resource
18h ago
Tau Ceti project resource
2 experts are actively discussing the implications.
1 expert
0 communities
1 sources clustered
“when my agents worked on github.com/gaearon/conw..., the vast majority of the work was also formalizing papers "underneath" it. so this depends on whether someone needed those pieces before or not. the good news is that we'll get more and more reusable piec…”
2 experts discussed this · 4 posts
Eli Sennesh: Didn't they say they actually spent most of that LOC formalizing a bunch of intermediate lemmas that nobody had ever formalized before?
dan: isn't the human proof also very long? and it "gets to" use any established mathematics (which also "gets to" use any established mathematics before it), whereas in Lean you actually have to prove e…
dan: when my agents worked on github.com/gaearon/conw..., the vast majority of the work was also formalizing papers "underneath" it. so this depends on whether someone needed those pieces before or not.…
Open the full discussion →
New
AI field signal
Analysis
5h ago
AI code-shaped world essay
2 experts are actively discussing the implications.
1 expert
0 communities
1 sources clustered
“Using the quote below as an excuse to re-share my recent essay: davidpreichert.substack.com/p/ai-and-a-c... This is another attempt at conveying intuitions about AI agents through concrete examples; from modelling historical houses and fixing crashes to bui…”
2 experts discussed this · 3 posts
david-p-reichert.bsky.social: A follow-up to my last substack about agents vs chatbots. This time I'm showing some more down-to-earth examples of agents in action, as opposed to machine learning. Anything from modelling histori…
david-p-reichert.bsky.social: The goal remains the same: convey to people who haven't been exposed to recent developments what is now becoming possible, so that we can take AI seriously and figure out what this all means for th…
Benjamin Riley: Interesting read. Re "what if more of the rest of the world shapes itself around coding?", given the mounting stories about AI agents exploiting cybersecurity vulnerabilities, I wonder (ok suspect)…
Open the full discussion →
New
AI field signal
Analysis
18h ago
Webster Massachusetts personal essay
2 experts are actively discussing the implications.
1 expert
1 community
1 sources clustered
“I'm having Astra slowly simulate driving in my hometown, will continue to get better over the month as my usage limits reset andymasley.com/town/”
2 experts discussed this · 3 posts
Andy Masley: I'm having Astra slowly simulate driving in my hometown, will continue to get better over the month as my usage limits reset andymasley.com/town/
Andy Masley: Yup slowly improving it
Grace: You should collaborate with @cee.wtf to make a zombie mode where you have to escape a crowd of angry posters
Open the full discussion →
Aspen Digital AI policy resource
2 experts are actively discussing the implications.
1 expert
1 community
1 sources clustered
“I still think the "race" framing is problematicccc, but I think there are cool ways to work with it www.aspendigital.org/report/strat...”
2 experts discussed this · 3 posts
brennan: this seems to be what I'm seeing (although I'm canadian, so maybe that itself is part of it) the hyperonline leftists are a bubble of anti-AI vitriol, but everyone I know socially (normie liberal t…
B! 🐝 Cavello (they/them): I still think the "race" framing is problematicccc, but I think there are cool ways to work with it www.aspendigital.org/report/strat...
B! 🐝 Cavello (they/them): I won't claim to be very steeped in leftism, but I think the retreat to copyright and property law is an interesting thing playing out over what I would consider an information commons. I think tha…
Open the full discussion →
