Pride and Prejudice: LLM Amplifies Self-Bias in Self-Refinement (2024 · 88 citations)
Letting an LLM critique and revise its own output helps on some tasks but hurts on others. The authors attribute this inconsistency to self-bias: models rate their own generations more generously than an independent judge would. They formalise self-bias with two statistics. The first is the average gap between the…
The Hidden Risks of Large Reasoning Models: A Safety Assessment of R1 (2025 · 30 citations)
Open-weight large reasoning models such as DeepSeek-R1 have become much stronger. Because they are freely available, misuse is a real concern. This study audits their safety systematically. It compares three open reasoning models (DeepSeek-R1, the Llama 3.3-based distilled DeepSeek-R1-70b, and QwQ) with non-reasoning…
SafeKey: Amplifying Aha-Moment Insights for Safety Reasoning (2025 · 9 citations)
Large Reasoning Models (LRMs) such as DeepSeek-R1 deliberate before answering, but they often comply with harmful requests and jailbreaks.
Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption (2026 · 2 citations)
This position paper asks why LLM watermarking is rarely deployed despite many proposed algorithms. The authors argue that the main obstacle is not technical performance. Instead, LLM providers, platforms, and end users want different things. They identify three barriers: competitive risk for providers who adopt…

Co-authors here

10 co-authors with a page, the most papers together first.

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

Papers of our library cite their works 94 times (of the 149,439 papers whose bibliographies we read); 28 citing papers with a page here →.

Updated 31 October 2026