"GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior" by Axel Højmark, Ezra Newman
LessWrong (Curated & Popular)Show Notes
Authors: Axel Højmark*, Ezra Newman* (*Equal contribution) While reviewing transcripts from standard behavioral evaluations of GLM-5.3, we noticed that the model occasionally reasons, unprompted, about how its actions will shape future models. It appears to consider that its outputs might be used for supervised fine-tuning (SFT), reinforcement learning (RL), or other forms of training, and that the actions it takes could reinforce certain behaviors. For example, GLM-5.3 sometimes reasons t...
More from LessWrong (Curated & Popular)
Ice cream There's a particular kind of listening that kids (and adults) sometimes do. Daddy is saying stuff—his mouth is moving and making noise—and he's talking about what I ate for lunch, and something about other times I've eaten ice cream, and something about rules and this and that and what Mommy would think and something about bedtime—but I am not so interested in that. What interests me is the punchline "...and therefore you can't have ice cream.". I'm only incidentally interested ...
Cross-posted from Sören Mindermann's substack (I am not Sören). Should be of general interest to the LW community. Next month, I am moving to Beijing and joining Tsinghua University's College of AI as an assistant professor working on frontier AI safety, especially loss of control. I have also started co-directing the new Beijing AI Safety Hub (BASH) in Tsinghua, and I continue my Cambridge work. I believe I can have a large impact there compared to strong baseline options. This post ex...
Why LM Psychology? One of the great ironies of modern language models is that we create them, we know exactly what computations constitute them, and yet we do not really understand them. More precisely, we do not understand them on a level of abstraction that lets us reliably predict how their behaviour will generalise: there's a trivial sense in which we can predict how they will behave in specific new situations — we can just run the computations — but we can’t answer the broader questio...
As our understanding of mechanistic interpretability and steering techniques increase, for both LLMs and uploaded connectomes, there is a corresponding "hellish capabilities" increase. In layman's terms, we get better at torturing the digital minds we create. Our brief history with open source technologies such as MaleCNS or the LLM pain steering repo shows that torture is often one of the first use cases pursued when the public is given access to a digital mind perceived as capable of suf...
This is a link post. Announcing www.huggingface.wtf, a site that explains: "Hugging Face. Wtf?" I wrote the most accessible-but-complete explanation I could. Along the way, I made a nice AI Quotes database that I think will be useful for people who already understand the Hugging Face incident but want to search everything the agents said. It really seems like humanity shouldn't need a second warning shot, after the Hugging Face Incident. Really, OpenAI's agents did a remarkable job highli...