11 papers
Can AI agents conduct open-ended AI research? Early evidence from two case studies
Peter Kirgis, Sayash Kapoor, Andrew Schwartz +21
The paper evaluates whether current AI agents can independently conduct open‑ended AI research by having them attempt to solve the central questions of two unpublished NeurIPS subm…
How Inference Compute Shapes Frontier LLM Evaluation
Jessica McFadyen, Ole Jorgensen, Harry Coppock +2
The paper studies how the amount of compute allocated during inference (e.g., token budget, repeated attempts) affects the performance of frontier large language models on challeng…
Open-World Evaluations for Measuring Frontier AI Capabilities
Sayash Kapoor, Peter Kirgis, Andrew Schwartz +15
Benchmark-based evaluation remains important for tracking frontier AI progress. But it can both overstate and understate deployed capability because it privileges tasks that can be…
Log analysis is necessary for credible evaluation of AI agents
Peter Kirgis, Sayash Kapoor, Stephan Rabanser +8
Agent benchmarks typically report only final outcomes: pass or fail. This threatens evaluation credibility in three ways. First, scores may be inflated or deflated by shortcuts and…
Ask don't tell: Reducing sycophancy in large language models
Magda Dubois, Cozmin Ududec, Christopher Summerfield +1
Sycophancy, the tendency of large language models to favour user-affirming responses over critical engagement, has been identified as an alignment failure, particularly in high-sta…
Seven simple steps for log analysis in AI systems
Magda Dubois, Ekin Zorer, Maia Hamin +17
AI systems produce large volumes of logs as they interact with tools and users. Analysing these logs can help understand model capabilities, propensities, and behaviours, or assess…