This research explores methods to protect Qwen3-8b from GCG-based persona jailbreaks by using linear direction steering. The study involves training linear probes to predict if an answer will adopt an assistant persona, with experiments focusing on steering through these probes to balance personas. The work builds on previous findings about LLMs randomly adopting personas under specific triggers.
The chart will appear after repeat observations. The current metric comes from the source.
8 source pointsReal observations only. History before source connection is not reconstructed.
This approach could help mitigate safety risks associated with LLMs adopting unintended personas under specific triggers, enhancing model reliability and control.
The author argues that while companies like OpenAI and Anthropic are slowing down RL training for safety, the science of loss-of-control risk in AI is still developing. There are no standardized methods to measure or verify safety claims, making it difficult to assess whether AI systems might undermine human control. The author suggests that AI companies should focus on improving their own safety practices rather than relying on third-party evaluations.
2 days agoLessWrongThe post discusses the tension between AI safety and capabilities research, arguing that focusing solely on safety can limit the impact of research. It examines the history of interpretability research and suggests that prioritizing safety over capabilities may hinder progress. The author proposes strategies for conducting alignment research without contributing to capabilities development.
4 days agoLessWrongThe post discusses the risks of AI in 2026, focusing on the dangers posed by a single institution (the "Frontier AI Company") that has the potential to create highly powerful and self-replicating entities. It argues that separating the institutional and financial aspects of such companies could mitigate most AI risks. The author suggests that using ASICs could help create a productive AI industry without the risks associated with a single entity controlling both the technology and financial incentives.
2 days agoLessWrongThe author expresses concerns about reinforcement learning (RL) from theoretical, practical, and future perspectives, highlighting risks of misalignment and potential negative behaviors in AI systems. They suggest strategies to mitigate these risks by reducing RL use, improving RL practices, and aligning incentives.
5 days agoLessWrongThe paper discusses the potential risks of latent reasoning architectures, which could reduce the effectiveness of Chain of Thought (CoT) as a tool for understanding AI systems. It highlights that such architectures might allow AI models to reason extensively in latent states rather than through text-based CoT, making oversight more challenging. The paper also mentions examples like COCONUT and full-bandwidth transformers that could enable this shift.
5 days agoLessWrongThe paper explores the concept of an AI sanctuary as a potential third option for rogue AIs, beyond criminal activity or shutdown, to address adverse selection pressures that may push rogue AIs toward criminal behavior. It discusses the possible benefits and risks of such a sanctuary, including its impact on AI alignment and information gathering, while acknowledging the exploratory nature of the proposal.
7 hours ago