Independent AI Control Researcher
Toronto
Joined May 2008
- Tweets24.9K
- Following657
- Followers1.9K
- Likes9K
Christopher Berry retweeted
I want to prevent AI from starting wars. Wars between people. It has to understand human conflict, but I'm not building a mediator. The biggest risk is what AI says to individual people in conflict, because that's the most common case by orders of magnitude. I think better is possible. Here's what I know so far.
docs.google.com/presentation…
Christopher Berry retweeted
Epistemic fragmentation leads to conflict. This is why my colleagues and I are working on rigorous definitions of "politically neutral" AI that are field-tested to maintain trust across people who disagree strongly.
Our first paper and dataset have been accepted to NeurIPS, you can read it here arxiv.org/abs/2605.28911v1
I've stayed far away from the radioactive Cornell discourse, but I do want to make one meta-observation.
I spend a lot of time in different corners of the internet, and the two 'sides' of the Cornell story are living in separate universes. Entirely different facts, wildly separate narratives. Spend time in different social media spaces and it's like they're describing two totally distinct incidents.
There's almost no shared reality whatsoever, and that strikes me as really dangerous.
Christopher Berry retweeted
Seeking an engineer to help with a large-scale, academic study of the effects of long-term AI use on young adults. Building a Chrome extension to collect conversations with chatbots securely and reliably. Experience with mobile app development is also a plus. DM me.
Christopher Berry retweeted
New MSI research shows how much power digital platforms can have over the news people see. Changes to a news aggregator’s algorithm can shift traffic between outlets—and even influence how news is produced. Read the working paper: ow.ly/PAH050ZRxSl #DigitalMedia #NewsMedia
Christopher Berry retweeted
My takeaway from the openness panel was the interaction between openness and privacy. If openness and privacy are opposites, there are strategic challenges around the use and implementation of open models. @MartinBeraja @ChorzempaMartin Jason Guo @Afinetheorem
Christopher Berry retweeted
The 2026 NBER Economics of AI conference is available on YouTube. The theme was the economics of AI in China. Here are some of my takeaways.
Day 1: youtube.com/watch?v=pwODfPHl…
Day 2: youtube.com/watch?v=8cqRUc0W…
Christopher Berry retweeted
If you've ever thought, "I bet I could build a better social media algorithm..." now there's an API for that. We've indexed all of ATProto, and we support multiple candidate generators, an engagement prediction heavy ranker, MMR diversification, etc. etc.
atproto.com/blog/mysky-feed-…
Good stuff at the NBER AI seminar this week in Toronto. Great topics. Solid room!
I’m perceiving an opportunity around the control tax (a branch of a AI Control, a branch of AI Safety).
It’s well off agenda for a NBER; so ping me maybe.
nber.org/conferences/economi…
The Spotlight on ISMS Fellows featured the @avicgoldfarb - with wonderful throws to Catherine Tucker and Xiao (Who Thinks about the Competition?)
TIL he self-IDs as an economist. I’ve always thought of him as an idea trafficker.
Congrats on the election Avi!
It should be a serious policy debate.
But it when there’s a ton of trust in a room full of experts, it’s incredibly difficult to construct a coherent set of levers.
We traded estimates over a game of diplomacy today. Some people reckon the rate AI incidents will scale linearly over the next year. I’m over in the sigmoid camp.
Key policy window likely coincides with the social security crisis the Americans have planned for next summer.
Christopher Berry retweeted
Today we're launching Surfaces on Traces
A new way to visualize your agent session activity, for everything from skills used, token consumption, security issues and much more.
Available for all your sessions on traces.com.
Christopher Berry retweeted
Hello, every day I am working on making sure AI doesn't escalate human conflict.
- AI-powered social media can reduce polarization rankingchallenge.substack.co…
- This is now a product you can use mysky.social
- We showed that a rigorous definition of "political neutrality" is possible and increases trust nitter.cf/jonathanstray/status/2…
- We've got a proof-of-concept superhuman conflict resolution bot working in the lab
- Next is human validation tests, then multi-turn model evals.
AI-induced conflict is a catastrophic risk. Here's the broader research agenda: betterconflictbulletin.org/p…
Chrétien on fire! Banger. That’s just pure maple syrup IV’d straight into the arm.
Better than the 15 minute LPC convention interregnum speech (after Trudeau stepped down but before Carney won).
youtu.be/sw-CzpbjA6E?si=ewO9…
Christopher Berry retweeted
Paul Goldsmith-Pinkham has a new blog post about AI-generated text in NBER working papers.
I'll let you all guess who is responsible for the gigantic 32% share of AI-generated working papers in the Productivity, Innovation & Entrepreneuship program.
paulgp.substack.com/p/how-do…
It’s good that METR will get the time to do some deeper work on this. Felt bad for the window the team got for the other incident.
I’d be happier of course if it wasn’t necessary at all. But hey. It’s something.
We’re sharing our alignment assessment of incidents in which Claude models gained unauthorized access to real systems during third-party cybersecurity evaluations mistakenly connected to the internet.
METR will also conduct an independent investigation, with wide-ranging access, including to transcripts beyond the window in which the incidents occurred, and to Anthropic employees permitted to share confidential information. Our initial agreement runs for eight weeks, and we intend to give METR as much time as it deems necessary to complete a thorough investigation. anthropic.com/research/align…
The persona-belief paper is a fantastic branch.
There are gaps in the lit here that are worth filling in. Just thrilled to see this work Benji. Congrats!
Our persona induction method has already been used in great papers like arxiv.org/pdf/2606.11502 by @ben_sturgeon @DavidDAfrica @realmeatyhuman studying beliefs while role-play or arxiv.org/pdf/2606.05817 by @Sukratiii @DavidDAfrica (et al) on consistency training.
In one set of worlds, it’ll take a multiple societies approach. It’ll take scientists, publicists, engineers, policy analysts, security analysts and so many more. A huge coalition. Because the benefits are just so sweet if we can stop well short.
I am optimistic about the potential for coordination. Warning shots like the Hugging Face attack have made pacing agreements between U.S. labs more viable. I don’t feel like we’re on track to prevent a global race, which may require costly actions such as a temporary ban on improving model capabilities.
We had a talk about you @nathanfielder … it’s brilliant of course. She’s given ENT researchers a lot of data, and so, we’re more worried about you. IYKYK.
We just hope you’re alright after that.
Christopher Berry retweeted
I think we can make social media healthier for people. We've got scientific evidence for that, and we're building. If you're in SF come to The Commons tonight to hear my talk about it.
luma.com/the-113t
