2022 · AI safety
Toy Models of Superposition
Selected research in ai safety, scaling laws. Read the full paper, including the methods, experiments, and reported results.
Paper & contextChris Olah
Interpretability
Investigating the internal features and computations of networks.
Olah studies the features and computations inside neural networks. The selected work connects interpretability with model behavior and alignment, including toy models of superposition and experiments aimed at understanding what language models represent.
Selected work
10 papers
2022 · Alignment
Constitutional AI: Harmlessness from AI Feedback
Constitutional AI explores training a more helpful and harmless assistant with written principles and AI-generated feedback.
Paper & context2021 · AI safety
A General Language Assistant as a Laboratory for Alignment
Selected research in ai safety, scaling laws. Read the full paper, including the methods, experiments, and reported results.
Paper & context2025 · Interpretability
Auditing language models for hidden objectives
Selected research in interpretability. Read the full paper, including the methods, experiments, and reported results.
Paper & context2022 · AI safety
Discovering Language Model Behaviors with Model-Written Evaluations
Selected research in ai safety, scaling laws. Read the full paper, including the methods, experiments, and reported results.
Paper & context2023 · AI safety
The Capacity for Moral Self-Correction in Large Language Models
Selected research in ai safety, scaling laws. Read the full paper, including the methods, experiments, and reported results.
Paper & context2022 · AI safety
Measuring Progress on Scalable Oversight for Large Language Models
Selected research in ai safety, scaling laws. Read the full paper, including the methods, experiments, and reported results.
Paper & context2018 · Generative models
Is Generator Conditioning Causally Related to GAN Performance?
Selected research in generative models, interpretability. Read the full paper, including the methods, experiments, and reported results.
Paper & context2017 · Interpretability
Changing Model Behavior at Test-Time Using Reinforcement Learning
Selected research in interpretability. Read the full paper, including the methods, experiments, and reported results.
Paper & context2016 · Interpretability
Conditional Image Synthesis With Auxiliary Classifier GANs
Selected research in interpretability. Read the full paper, including the methods, experiments, and reported results.
Paper & contextAn independent editorial profile. Inclusion does not imply Council membership or endorsement. Research is collaborative; coauthorship does not imply sole credit.
Selection & sources