misalignment scaling with model intelligence, recursive self-improvement, CoT obfuscation generalization, ...
Also: mitigating AI-discovered 0-days, light alignment with a single neuron
Chain-of-thought obfuscation learned from output supervision can generalise to unseen tasks
“Chain-of-thought (CoT) reasoning provides a significant performance uplift to LLMs by enabling planning, exploration, and deliberation of their actions. CoT is also a powerful tool for monitoring the behaviours of these agents: when faithful, they offer interpretations of the model’s decision making process, and an early warning sign for dangerous behaviours. However, optimisation pressures placed on the CoT may cause the model to obfuscate reasoning traces, losing this beneficial property. We show that obfuscation can generalise across tasks; models that learn to obfuscate reasoning involving reward hacking (e.g. accessing and utilising leaked information) generalise both the reward hacking behaviour and its obfuscation in CoT to unseen reward hacking settings. Most worryingly, we show that obfuscation of CoT reasoning, and its generalisation across tasks, also follows when we penalise only the model’s final actions after closing its CoT. Our findings suggest that current practices of penalising harmful generations may inadvertently lead to a reduction in the broader monitorability of LLMs in unpredictable ways.”
Light Alignment Improves LLM Safety via Model Self-Reflection with a Single Neuron
“The safety of large language models (LLMs) has increasingly emerged as a fundamental aspect of their development. Existing safety alignment for LLMs is predominantly achieved through post-training methods, which are computationally expensive and often fail to generalize well across different models. A small number of lightweight alignment approaches either rely heavily on prior-computed safety injections or depend excessively on the model’s own capabilities, resulting in limited generalization and degraded efficiency and usability during generation. In this work, we propose a safety-aware decoding method that requires only low-cost training of an expert model and employs a single neuron as a gating mechanism. By effectively balancing the model’s intrinsic capabilities with external guidance, our approach simultaneously preserves utility and enhances output safety. It demonstrates clear advantages in training overhead and generalization across model scales, offering a new perspective on lightweight alignment for the safe and practical deployment of large language models.”
The Hot Mess of AI: How Does Misalignment Scale with Model Intelligence and Task Complexity?
“When AI systems fail, will they fail by systematically pursuing the wrong goals, or by being a hot mess? We decompose the errors of frontier reasoning models into bias (systematic) and variance (incoherent) components and find that, as tasks get harder and reasoning gets longer, model failures become increasingly dominated by incoherence rather than systematic misalignment. This suggests that future AI failures may look more like industrial accidents than coherent pursuit of a goal we did not train them to pursue.”
On Recursive Self-Improvement (Part I)
“This year, the automation of AI research and engineering will begin in earnest. In addition to creating at least a step-change improvement in AI progress from its already rapid pace, this could change the dynamics of AI competition, alter AI geopolitics, and much more.”
Evaluating and mitigating the growing risk of LLM-discovered 0-days
“AI models can now find high-severity vulnerabilities at scale. Our view is this is a moment to move quickly—to empower defenders and secure as much code as possible while the window exists.”