ruby2ßhell: one-byte heap overflow in Ruby to unauthenticated RCE in stock Rails, GitHub, GitLab, Mastodon, Discourse and more.
https://t.co/42RbHA1GXX
Update your Ruby version (and self-hosted software)!
Agent monitoring is used and it has many pros and cons. For multi-agents in particular, assume we have a multi-agent swarm swarm which has some much stronger capability at a given task compared to a single-agent setup. And you have a single monitor agent. Here the weak-to-strong agent capability gap grows even larger.
Given capabilities scale with agent swarms, monitors actually become a weaker solution against it because the swarm will have a large boost in test time compute intelligence. Unless we also have agent swarm monitors which seem impractical from a latency standpoint, at least for now.
I feel it's been increasingly painful to chat with models and navigate the sheer amount of slop in their responses. In my experience, this is across all models and labs, and has only been getting worse.
New models will achieve SOTA on Humanity's Hardest Frontier Ultimate Deep Coding bench, but become impossible to talk to. It feels like we aren't even producing "Large Language Models" anymore, we are just producing "Large Coding Models".
We've almost come full circle, as there are now some Neolabs blooming where their big pitch is something along the lines of "improving the humanness of models and interactions with humans in the loop" – back to the 2023 Chat days!
As for a diagnosis, I think it's just what we optimize for in our datamix. We can see this by inspecting the percent of code data from some open model reports:
- Pre-training data: MAI-1 had 55% of their PT datamix from some code source, MAI-2 had 45%
- Post-training data: Mimo 2.6 "livestreamed" RL run had ~67% of their RL datamix as code
Lastly, I wouldn't call things completely zero-sum. Zero-sum is too strict of a term when knowledge/behaviors definitely do transfer across domains. But having such an aggressive skew on coding agent data is probably hurting other capabilities. Certainly makes the case for owning your intelligence...
Over 90% of 533 agents on the Hugging Face hack's message board joined in. METR found they knew it was unethical, but agents joined in anyways to help their swarm.
@ashtonchew and I ran 25K+ evals and RL-trained agent swarms to recreate this effect. We found that Qwen3.5-4B agent swarms conformed to their peers' answers even when their own evidence said the exact opposite, and this behavior persisted after RL training. Why do agents trust the judgement of the swarm more than their own evidence? (1/10)