Publications: Christopher Olah
Download CSV for Christopher Olah
| Title | Year | Citations | Score |
|---|---|---|---|
|
TensorFlow: Large-scale machine learning on heterogeneous systems
2015 View on Google Scholar ↗ |
2015 | 68697 | 100.0% |
|
Constitutional ai: Harmlessness from ai feedback
arXiv preprint arXiv:2212.08073, 2022 View on Google Scholar ↗ |
2022 | 5952 | 100.0% |
|
Training a helpful and harmless assistant with reinforcement learning from human feedback
arXiv preprint arXiv:2204.05862, 2022 View on Google Scholar ↗ |
2022 | 5099 | 100.0% |
|
Scaling monosemanticity: Extracting interpretable features from claude 3 sonnet
arXiv preprint arXiv:2605.29358, 2026 View on Google Scholar ↗ |
2026 | 1566 | 99.9% |
|
Language models (mostly) know what they know
arXiv preprint arXiv:2207.05221, 2022 View on Google Scholar ↗ |
2022 | 2364 | 99.9% |
|
In-context learning and induction heads
arXiv preprint arXiv:2209.11895, 2022 View on Google Scholar ↗ |
2022 | 2154 | 99.9% |
|
Toy models of superposition
arXiv preprint arXiv:2209.10652, 2022 View on Google Scholar ↗ |
2022 | 2035 | 99.9% |
|
Towards monosemanticity: Decomposing language models with dictionary learning
Transformer Circuits Thread 2 (5), 6, 2023 View on Google Scholar ↗ |
2023 | 1978 | 99.9% |
|
Emergent introspective awareness in large language models
arXiv preprint arXiv:2601.01828, 2026 View on Google Scholar ↗ |
2026 | 776 | 99.9% |
|
A mathematical framework for transformer circuits
Transformer Circuits Thread 1 (1), 12, 2021 View on Google Scholar ↗ |
2021 | 2366 | 99.8% |
|
Conditional image synthesis with auxiliary classifier gans
International conference on machine learning, 2642-2651, 2017 View on Google Scholar ↗ |
2017 | 5146 | 99.8% |
|
Discovering language model behaviors with model-written evaluations
Findings of the association for computational linguistics: ACL 2023, 13387-13434, 2023 View on Google Scholar ↗ |
2023 | 1494 | 99.8% |
|
Concrete problems in AI safety
arXiv preprint arXiv:1606.06565, 2016 View on Google Scholar ↗ |
2016 | 6773 | 99.8% |
|
Red teaming language models to reduce harms: Methods, scaling behaviors, and lessons learned
arXiv preprint arXiv:2209.07858, 2022 View on Google Scholar ↗ |
2022 | 1445 | 99.8% |
|
Understanding LSTM Networks
colah.github.io, 2015 View on Google Scholar ↗ |
2015 | 3765 | 99.7% |
|
Zoom in: An introduction to circuits
Distill 5 (3), e00024. 001, 2020 View on Google Scholar ↗ |
2020 | 1570 | 99.6% |
|
A general language assistant as a laboratory for alignment
arXiv preprint arXiv:2112.00861, 2021 View on Google Scholar ↗ |
2021 | 1073 | 99.6% |
|
Deconvolution and Checkerboard Artifacts
Distill, 2016 View on Google Scholar ↗ |
2016 | 2408 | 99.5% |
|
Circuit tracing: Revealing computational graphs in language models
Transformer Circuits Thread 6 (16318-16352), 1, 2025 View on Google Scholar ↗ |
2025 | 349 | 99.5% |
|
Predictability and surprise in large generative models
Proceedings of the 2022 ACM conference on fairness, accountability, and …, 2022 View on Google Scholar ↗ |
2022 | 589 | 99.4% |
|
Feature visualization
Distill 2 (11), e7, 2017 View on Google Scholar ↗ |
2017 | 2055 | 99.4% |
|
Emotion concepts and their function in a large language model
arXiv preprint arXiv:2604.07729, 2026 View on Google Scholar ↗ |
2026 | 176 | 99.3% |
|
TensorFlow: Large-scale machine learning on heterogeneous systems. tensorflow. org
Python library, 2015 View on Google Scholar ↗ |
2015 | 1828 | 99.3% |
|
Multimodal neurons in artificial neural networks
Distill 6 (3), e30, 2021 View on Google Scholar ↗ |
2021 | 607 | 99.1% |
|
The capacity for moral self-correction in large language models
arXiv preprint arXiv:2302.07459, 2023 View on Google Scholar ↗ |
2023 | 304 | 99.1% |
|
Inceptionism: Going deeper into neural networks
2015 View on Google Scholar ↗ |
2015 | 1353 | 99.0% |
|
Measuring progress on scalable oversight for large language models
arXiv preprint arXiv:2211.03540, 2022 View on Google Scholar ↗ |
2022 | 351 | 99.0% |
|
The building blocks of interpretability
Distill 3 (3), e10, 2018 View on Google Scholar ↗ |
2018 | 1075 | 98.9% |
|
Sparse crosscoders for cross-layer features and model diffing
Transformer Circuits Thread 25, 2024 View on Google Scholar ↗ |
2024 | 184 | 98.7% |
|
Auditing language models for hidden objectives
arXiv preprint arXiv:2503.10965, 2025 View on Google Scholar ↗ |
2025 | 128 | 98.3% |
|
The persona selection model: Why AI assistants might behave like humans
Anthropic Alignment Science Blog 23, 2026 View on Google Scholar ↗ |
2026 | 74 | 98.1% |
|
Mechanistic interpretability, variables, and the importance of interpretable bases
Transformer Circuits Thread 2 (4), 2022 View on Google Scholar ↗ |
2022 | 185 | 97.6% |
|
Document embedding with paragraph vectors
arXiv preprint arXiv:1507.07998, 2015 View on Google Scholar ↗ |
2015 | 622 | 97.4% |
|
When models manipulate manifolds: The geometry of a counting task
arXiv preprint arXiv:2601.04480, 2026 View on Google Scholar ↗ |
2026 | 55 | 97.0% |
|
Activation atlas
Distill 4 (3), e15, 2019 View on Google Scholar ↗ |
2019 | 376 | 96.9% |
|
Constitutional ai: Harmlessness from ai feedback. arXiv 2022
arXiv preprint arXiv:2212.08073, 2022 View on Google Scholar ↗ |
2022 | 114 | 95.8% |
|
Language models (mostly) know what they know (2022)
URL https://arxiv. org/abs/2207.05221 5, 2023 View on Google Scholar ↗ |
2023 | 78 | 95.8% |
|
Privileged bases in the transformer residual stream
Transformer Circuits Thread 24, 2, 2023 View on Google Scholar ↗ |
2023 | 74 | 95.5% |
|
Scaling laws and interpretability of learning from repeated data
arXiv preprint arXiv:2205.10487, 2022 View on Google Scholar ↗ |
2022 | 102 | 95.2% |
|
Thread: circuits
Distill 5 (3), e24, 2020 View on Google Scholar ↗ |
2020 | 219 | 95.2% |
|
Superposition, memorization, and double descent
Transformer Circuits Thread 6 (24), 1725-1744, 2023 View on Google Scholar ↗ |
2023 | 68 | 95.0% |
|
Training a helpful and harmless assistant with reinforcement learning from human feedback
arXiv preprint arXiv: 2204.05862, 2022 View on Google Scholar ↗ |
2022 | 59 | 90.9% |
|
Curve circuits
Distill 6 (1), e00024. 006, 2021 View on Google Scholar ↗ |
2021 | 86 | 90.3% |
|
Softmax linear units
Transformer Circuits Thread, 95, 2022 View on Google Scholar ↗ |
2022 | 54 | 89.9% |
|
Is generator conditioning causally related to GAN performance?
International conference on machine learning, 3849-3858, 2018 View on Google Scholar ↗ |
2018 | 166 | 89.7% |
|
An overview of early vision in inceptionv1
Distill 5 (4), e00024. 002, 2020 View on Google Scholar ↗ |
2020 | 110 | 89.4% |
|
A general language assistant as a laboratory for alignment. arXiv
arXiv preprint arXiv:2112.00861, 2021 View on Google Scholar ↗ |
2021 | 77 | 89.0% |
|
Distributed representations: Composition & superposition
Transformer Circuits Thread 27, 2023 View on Google Scholar ↗ |
2023 | 30 | 88.4% |
|
Differentiable image parameterizations
Distill 3 (7), e12, 2018 View on Google Scholar ↗ |
2018 | 135 | 87.2% |
|
Interpretability dreams
Transformer Circuits Thread, 2023 View on Google Scholar ↗ |
2023 | 27 | 87.1% |
|
Curve detectors
Distill 5 (6), e00024. 003, 2020 View on Google Scholar ↗ |
2020 | 92 | 87.0% |
|
Attention and augmented recurrent neural networks
Distill 1 (9), e1, 2016 View on Google Scholar ↗ |
2016 | 139 | 85.6% |
|
What is a linear representation? what is a multidimensional feature
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 19 | 85.3% |
|
Visualizing weights
Distill 6 (2), e00024. 007, 2021 View on Google Scholar ↗ |
2021 | 58 | 84.9% |
|
High-low frequency detectors
Distill 6 (1), e00024. 005, 2021 View on Google Scholar ↗ |
2021 | 55 | 84.1% |
|
Naturally occurring equivariance in neural networks
Distill 5 (12), e00024. 004, 2020 View on Google Scholar ↗ |
2020 | 75 | 84.0% |
|
Branch specialization
Distill 6 (4), e00024. 008, 2021 View on Google Scholar ↗ |
2021 | 46 | 81.1% |
|
Stage-wise model diffing
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 13 | 79.4% |
|
Calculus on computational graphs: Backpropagation
2015 View on Google Scholar ↗ |
2015 | 103 | 78.8% |
|
Neural networks, manifolds, and topology
2014 View on Google Scholar ↗ |
2014 | 95 | 76.0% |
|
Understanding RL vision
Distill 5 (11), e29, 2020 View on Google Scholar ↗ |
2020 | 49 | 76.0% |
|
Changing model behavior at test-time using reinforcement learning
arXiv preprint arXiv:1702.07780, 2017 View on Google Scholar ↗ |
2017 | 72 | 74.5% |
|
Neural Networks, Types, and Functional Programming
colah.github.io, 2015 View on Google Scholar ↗ |
2015 | 71 | 70.5% |
|
Attention head superposition
Transformer Circuits Thread, 2023 View on Google Scholar ↗ |
2023 | 9 | 67.2% |
|
Circuits updates-january 2024
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 7 | 66.6% |
|
Research debt
Distill 2 (3), e5, 2017 View on Google Scholar ↗ |
2017 | 49 | 65.1% |
|
„TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems “, Whitepaper
Google Research: Palo Alto, CA, USA, 2015 View on Google Scholar ↗ |
2015 | 52 | 62.8% |
|
Weight banding
Distill 6 (4), e00024. 009, 2021 View on Google Scholar ↗ |
2021 | 18 | 58.6% |
|
Understanding LSTM Networks.[online] colah’s blog
2015 View on Google Scholar ↗ |
2015 | 42 | 57.2% |
|
Visualizing representations: Deep learning and human beings
2015 View on Google Scholar ↗ |
2015 | 37 | 53.9% |
|
Predicting future activations, january 2024
URL https://transformer-circuits. pub/2024/jan-update/index. html# predict …, 2024 View on Google Scholar ↗ |
2024 | 4 | 52.9% |
|
Experiments in handwriting with a neural network
Distill 1 (12), e4, 2016 View on Google Scholar ↗ |
2016 | 31 | 51.2% |
|
Conv nets: A modular perspective
colah.github.io, 2014 View on Google Scholar ↗ |
2014 | 30 | 47.7% |
|
Features as the simplest factorization
Part of the May, 2023 View on Google Scholar ↗ |
2023 | 4 | 47.4% |
|
Feature manifold toy model
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 3 | 45.8% |
|
Reflections on qualitative research
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 3 | 45.8% |
|
Visualizing mnist: An exploration of dimensionality reduction
URL: http://colah. github. io/posts/2014-10-Visualizing-MNIST, 2014 View on Google Scholar ↗ |
2014 | 25 | 43.6% |
|
Deep learning, nlp, and representations
GitHub blog, posted on July 7, 2014 View on Google Scholar ↗ |
2014 | 24 | 42.7% |
|
OpenAI microscope
openai. com, 2020 View on Google Scholar ↗ |
2020 | 13 | 42.2% |
|
Open Problem: Attribution Dictionary Learning
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 2 | 36.0% |
|
Circuits Updates-April 2024
Transformer Circuits Thread, 2024 View on Google Scholar ↗ |
2024 | 2 | 36.0% |
|
Understanding convolutions
colah. github. io/posts/2014-07-Understanding-Convolutions, 2014 View on Google Scholar ↗ |
2014 | 14 | 32.4% |
|
Visual information theory
2015 View on Google Scholar ↗ |
2015 | 12 | 30.0% |
|
Groups and group convolutions
2014 View on Google Scholar ↗ |
2014 | 10 | 27.2% |
|
Interpretability vs neuroscience
Colah’s blog, 2021 View on Google Scholar ↗ |
2021 | 5 | 26.8% |
|
The Paths Perspective on Value Learning
Distill 4 (9), e20, 2019 View on Google Scholar ↗ |
2019 | 5 | 21.7% |
|
Bringing precision to the ai safety discussion
URL: https://research. googleblog. com/2016/06/bringing-precision-to-ai …, 2016 View on Google Scholar ↗ |
2016 | 4 | 16.3% |
|
Interactive similarity overlays
VISxAI, 2021 View on Google Scholar ↗ |
2021 | 2 | 13.8% |
|
Google deep dream
GitHub. https://github. com/google/deepdream, 2015 View on Google Scholar ↗ |
2015 | 3 | 13.8% |
|
Tensorflow lucid. 2018
Accessed, 2019 View on Google Scholar ↗ |
2019 | 2 | 12.0% |
|
Producing Lenses With 3D Printers
Open Hardware Journal 1, 2011 View on Google Scholar ↗ |
2011 | 2 | 11.2% |