iiinova — where your digital self keeps evolving

Reposted by Brian Christian

Center for Human-Compatible AI @chai-berkeley.bsky.social · 02/10/2026

The NYT covers @brianchristian.bsky.social's recent COLM paper in their article on the effects of AI assistance on learning and motivation: www.nytimes.com/interactive/...

nytimes.com

Will A.I. Make Your Brain Lazy? Here’s What the New Research Actually Shows.

We’re starting to learn more about how the technology can change us.

021

Reposted by Brian Christian

Serena Booth @reniebird.bsky.social · 23/09/2026

www.lesswrong.com/posts/HsijSh... @bradknox.bsky.social, Brian Christian, and I wrote a thing. We argue that an unexamined cause of the Open AI / Hugging Face Attack was the Exploit Gym evaluation metric. We post this on Less Wrong as a plea to practitioners to design better metrics in the future.

lesswrong.com

An unexamined cause of the OpenAI Hugging Face hacking incident: its binary performance metric — LessWrong

We argue that a main cause of the OpenAI Hugging Face incident was overlooked: the overly simple evaluation metric in ExploitGym was misaligned. Furt…

2259

Reposted by Brian Christian

Rachit Dubey @rachitdubey.bsky.social · 07/04/2026

🚨New preprint and our results are rather concerning.. We find the "boiling frog" equivalent of AI use. Using large-scale RCTs, we provide *casual* evidence that AI assistance reduces persistence and hurts independent performance. And these effects emerge after just 10–15 minutes of AI use! 1/

261533684

Brian Christian @brianchristian.bsky.social · 11/02/2026

My friend and collaborator of 21 years - and my coauthor on Algorithms to Live By - Tom Griffiths has a book out this week on the story of computational cognitive science. If you enjoyed Algorithms to Live By you won't want to miss it. Highly recommended:

0111

Reposted by Brian Christian

Jess Thompson @tsonj.bsky.social · 09/02/2026

We audited one of the most critical pieces of modern AI alignment: reward models. We find consistent and persistent biases and trace them back to the pretraining stage, challenging the premises of common approaches to alignment based on finetuning on human preferences. Accepted at #ICLR2026

0112

Brian Christian @brianchristian.bsky.social · 10/02/2026

New Preprint with @matanmazor.bsky.social: Overcoming both bias and sycophancy requires LLMs to imagine not knowing something they know. Like humans, they struggle with this. But unlike humans, LLMs can do something remarkable: they can, quite simply, *ask their counterfactual selves*.

1104

Brian Christian @brianchristian.bsky.social · 04/02/2026

Reward models (RMs) are supposed to represent human values. But RMs are NOT blank slates – they inherit measurable biases from their base models that stubbornly persist through preference training. #ICLR2026 🧵

1187

Brian Christian @brianchristian.bsky.social · 07/07/2025

Wow! Honored and amazed that our reward models paper has resonated so strongly with the community. Grateful to my co-authors and inspired by all the excellent reward model work at FAccT this year - excited to see the space growing and intrigued to see where things are headed next.

070

Brian Christian @brianchristian.bsky.social · 23/06/2025

Reward models (RMs) are the moral compass of LLMs – but no one has x-rayed them at scale. We just ran the first exhaustive analysis of 10 leading RMs, and the results were...eye-opening. Wild disagreement, base-model imprint, identity-term bias, mere-exposure quirks & more: 🧵

1435