corpus.blog
56,966 blogs · [ { "id": "01a07d5d-6c02-7045-957d-7fa29517fc8c", "title": "What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospective", "url": "https://blog.eleuther.ai/aletheia-retrospective/", "published_at": "2026-08-25T16:35:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa295fce5e8", "title": "A Dynamical Model of AI Governability", "url": "https://blog.eleuther.ai/dynamical-models-of-ai-governability/", "published_at": "2026-07-13T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2966c2125", "title": "Early Indicators of Reward Hacking via Reasoning Interpolation", "url": "https://blog.eleuther.ai/reward-hacking-indicators/", "published_at": "2026-04-15T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa296e8b5e4", "title": "Reward Hacking Resarch Update", "url": "https://blog.eleuther.ai/reward_hacking/", "published_at": "2025-10-07T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29760ed2b", "title": "Pretraining Data Filtering for Open-Weight AI Safety", "url": "https://blog.eleuther.ai/deep-ignorance/", "published_at": "2025-08-12T20:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa297c8361b", "title": "Attention Probes", "url": "https://blog.eleuther.ai/attention-probes/", "published_at": "2025-08-01T15:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2984d9646", "title": "Research Update: Applications of Local Volume Measurement", "url": "https://blog.eleuther.ai/tyche-poser-comparison/", "published_at": "2025-06-23T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa298d48909", "title": "Studying inductive biases of random networks via local volumes", "url": "https://blog.eleuther.ai/inductive-bias/", "published_at": "2025-06-12T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2999046a1", "title": "The Common Pile v0.1", "url": "https://blog.eleuther.ai/common-pile/", "published_at": "2025-06-05T20:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa299d54cfc", "title": "Product Key Memory Sparse Coders", "url": "https://blog.eleuther.ai/pkm-coders/", "published_at": "2025-05-30T22:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29a7f86e8", "title": "SAEs trained on the same data don’t learn the same features", "url": "https://blog.eleuther.ai/sae_seed_similarity/", "published_at": "2024-12-12T16:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29a9b37ad", "title": "Partially rewriting an LLM in natural language", "url": "https://blog.eleuther.ai/generating-text-using-nl-to-simulate-activations/", "published_at": "2024-11-10T16:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29b0e7af8", "title": "Third-party evaluation to identify risks in LLMs’ training data", "url": "https://blog.eleuther.ai/third-party-evals/", "published_at": "2024-10-31T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29b4f7ab2", "title": "Mechanistic Anomaly Detection Research Update 2", "url": "https://blog.eleuther.ai/mad_research_update_2/", "published_at": "2024-10-14T05:39:43+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29bff949b", "title": "RLHF and RLAIF in GPT-NeoX", "url": "https://blog.eleuther.ai/rlhf-and-rlaif-in-gpt-neox/", "published_at": "2024-10-10T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29cd716b1", "title": "The Practitioner's Guide to the Maximal Update Parameterization", "url": "https://blog.eleuther.ai/mutransfer/", "published_at": "2024-09-19T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29d96c69a", "title": "Mechanistic Anomaly Detection Research Update", "url": "https://blog.eleuther.ai/mad_research_update/", "published_at": "2024-08-05T16:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29e2af26a", "title": "Open Source Automated Interpretability for Sparse Autoencoder Features", "url": "https://blog.eleuther.ai/autointerp/", "published_at": "2024-07-30T22:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29ecc67bb", "title": "Experiments in Weak-to-Strong Generalization", "url": "https://blog.eleuther.ai/weak-to-strong/", "published_at": "2024-06-14T11:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa29f4dca5f", "title": "Free Form Least-Squares Concept Erasure Without Oracle Concept Labels", "url": "https://blog.eleuther.ai/free-form-leace/", "published_at": "2024-06-13T16:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a00b8235", "title": "VINC-S: Closed-form Optionally-supervised Knowledge Elicitation with Paraphrase Invariance", "url": "https://blog.eleuther.ai/vincs/", "published_at": "2024-05-22T17:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a094dd79", "title": "Pile-T5", "url": "https://blog.eleuther.ai/pile-t5/", "published_at": "2024-04-14T17:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a0ac5462", "title": "Yi-34B, Llama 2, and common practices in LLM training: a fact check of the New York Times", "url": "https://blog.eleuther.ai/nyt-yi-34b-response/", "published_at": "2024-03-25T09:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a1293f93", "title": "The Foundation Model Development Cheatsheet", "url": "https://blog.eleuther.ai/fm-dev-cheatsheet/", "published_at": "2024-02-29T09:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a156f500", "title": "Least-Squares Concept Erasure with Oracle Concept Labels", "url": "https://blog.eleuther.ai/oracle-leace/", "published_at": "2023-12-19T22:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a1a2236e", "title": "Diff-in-Means Concept Editing is Worst-Case Optimal", "url": "https://blog.eleuther.ai/diff-in-means/", "published_at": "2023-12-11T22:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a2941e81", "title": "The third New England RLHF Hackers Hackathon", "url": "https://blog.eleuther.ai/nerh_3/", "published_at": "2023-11-26T15:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a331c223", "title": "Extending the RoPE", "url": "https://blog.eleuther.ai/yarn/", "published_at": "2023-11-13T22:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a3c2c8d9", "title": "How the Foundation Model Transparency Index Distorts Transparency", "url": "https://blog.eleuther.ai/fmti-critique/", "published_at": "2023-10-26T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a3effdfe", "title": "Llemma: An Open Language Model For Mathematics", "url": "https://blog.eleuther.ai/llemma/", "published_at": "2023-10-17T02:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a454b06a", "title": "The second New England RLHF Hackers Hackathon", "url": "https://blog.eleuther.ai/nerh_2/", "published_at": "2023-10-13T20:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a4d36eb6", "title": "Contributor Spotlight: Mohammad Aflah Khan", "url": "https://blog.eleuther.ai/contributor-spotlight-1/", "published_at": "2023-09-21T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a57855ad", "title": "The first New England RLHF Hackers Hackathon", "url": "https://blog.eleuther.ai/nerh_1/", "published_at": "2023-09-19T20:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a595374c", "title": "EleutherAI's Thoughts on the EU AI Act", "url": "https://blog.eleuther.ai/eu-aia/", "published_at": "2023-07-26T17:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a597930c", "title": "Minetester: A fully open RL environment built on Minetest", "url": "https://blog.eleuther.ai/minetester-intro/", "published_at": "2023-07-08T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a61a5721", "title": "🐶Safetensors audited as really safe and becoming the default", "url": "https://blog.eleuther.ai/safetensors-security-audit/", "published_at": "2023-05-23T01:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a6a4a278", "title": "Alignment Research @ EleutherAI", "url": "https://blog.eleuther.ai/alignment-eleuther/", "published_at": "2023-05-03T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a75f0a26", "title": "Transformer Math 101", "url": "https://blog.eleuther.ai/transformer-math/", "published_at": "2023-04-17T23:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a7f96f6a", "title": "Exploratory Analysis of TRLX RLHF Transformers with TransformerLens", "url": "https://blog.eleuther.ai/trlx-exploratory-analysis/", "published_at": "2023-04-02T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a8690a9e", "title": "EleutherAI Second Retrospective: The long version", "url": "https://blog.eleuther.ai/year-two-full/", "published_at": "2023-03-26T22:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a8f81cc3", "title": "The View from 30,000 Feet: Preface to the Second EleutherAI Retrospective", "url": "https://blog.eleuther.ai/year-two-preface/", "published_at": "2023-03-02T07:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a9a49532", "title": "Announcing GPT-NeoX-20B", "url": "https://blog.eleuther.ai/announcing-20b/", "published_at": "2022-02-02T16:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2a9b7e1a1", "title": "A Preliminary Exploration into Factored Cognition with Language Models", "url": "https://blog.eleuther.ai/factored-cognition/", "published_at": "2021-10-25T20:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2aa096451", "title": "Multiple Choice Normalization in LM Evaluation", "url": "https://blog.eleuther.ai/multiple-choice-normalization/", "published_at": "2021-10-11T15:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2aa135434", "title": "Downstream Evaluations of Rotary Position Embeddings", "url": "https://blog.eleuther.ai/rotary-embeddings-eval-harness/", "published_at": "2021-08-16T18:13:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2aa405bed", "title": "What A Long, Strange Trip It's Been: EleutherAI One Year Retrospective", "url": "https://blog.eleuther.ai/year-one/", "published_at": "2021-07-08T00:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2aa7ac1b5", "title": "Why Release a Large Language Model?", "url": "https://blog.eleuther.ai/why-release-a-large-language-model/", "published_at": "2021-06-02T21:30:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2ab253aa7", "title": "On the Sizes of OpenAI API Models", "url": "https://blog.eleuther.ai/gpt3-model-sizes/", "published_at": "2021-05-24T20:00:03+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2abaf4b5d", "title": "Evaluating Different Fewshot Description Prompts on GPT-3", "url": "https://blog.eleuther.ai/prompts-gpt-fewshot/", "published_at": "2021-05-24T20:00:02+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2ac9c02a5", "title": "Finetuning Models on Downstream Tasks", "url": "https://blog.eleuther.ai/tuning-on-eval-harness/", "published_at": "2021-05-24T20:00:01+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2acbd5763", "title": "Activation Function Ablation", "url": "https://blog.eleuther.ai/activation-fns/", "published_at": "2021-05-24T20:00:00+00:00" }, { "id": "01a07d5d-6c02-7045-957d-7fa2ad4a2862", "title": "Rotary Embeddings: A Relative Revolution", "url": "https://blog.eleuther.ai/rotary-embeddings/", "published_at": "2021-04-21T01:00:00+00:00" } ] posts Claim your blog
Back to EleutherAI
Blog · corpus.blog/blogs/eleuther.ai/posts

EleutherAI

eleuther.ai

2026

What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospectiveoriginal ↗
25 Aug 2026
A Dynamical Model of AI Governabilityoriginal ↗
13 Jul 2026
Early Indicators of Reward Hacking via Reasoning Interpolationoriginal ↗
15 Apr 2026

2025

Reward Hacking Resarch Updateoriginal ↗
7 Oct 2025
Pretraining Data Filtering for Open-Weight AI Safetyoriginal ↗
12 Aug 2025
Attention Probesoriginal ↗
1 Aug 2025
Research Update: Applications of Local Volume Measurementoriginal ↗
23 Jun 2025
Studying inductive biases of random networks via local volumesoriginal ↗
12 Jun 2025
The Common Pile v0.1original ↗
5 Jun 2025
Product Key Memory Sparse Codersoriginal ↗
30 May 2025

2024

SAEs trained on the same data don’t learn the same featuresoriginal ↗
12 Dec 2024
Partially rewriting an LLM in natural languageoriginal ↗
10 Nov 2024
Third-party evaluation to identify risks in LLMs’ training dataoriginal ↗
31 Oct 2024
Mechanistic Anomaly Detection Research Update 2original ↗
14 Oct 2024
RLHF and RLAIF in GPT-NeoXoriginal ↗
10 Oct 2024
The Practitioner's Guide to the Maximal Update Parameterizationoriginal ↗
19 Sept 2024
Mechanistic Anomaly Detection Research Updateoriginal ↗
5 Aug 2024
Open Source Automated Interpretability for Sparse Autoencoder Featuresoriginal ↗
30 Jul 2024
Experiments in Weak-to-Strong Generalizationoriginal ↗
14 Jun 2024
Free Form Least-Squares Concept Erasure Without Oracle Concept Labelsoriginal ↗
13 Jun 2024
VINC-S: Closed-form Optionally-supervised Knowledge Elicitation with Paraphrase Invarianceoriginal ↗
22 May 2024
Pile-T5original ↗
14 Apr 2024
Yi-34B, Llama 2, and common practices in LLM training: a fact check of the New York Timesoriginal ↗
25 Mar 2024
The Foundation Model Development Cheatsheetoriginal ↗
29 Feb 2024

2023

Least-Squares Concept Erasure with Oracle Concept Labelsoriginal ↗
19 Dec 2023
Diff-in-Means Concept Editing is Worst-Case Optimaloriginal ↗
11 Dec 2023
The third New England RLHF Hackers Hackathonoriginal ↗
26 Nov 2023
Extending the RoPEoriginal ↗
13 Nov 2023
How the Foundation Model Transparency Index Distorts Transparencyoriginal ↗
26 Oct 2023
Llemma: An Open Language Model For Mathematicsoriginal ↗
17 Oct 2023
The second New England RLHF Hackers Hackathonoriginal ↗
13 Oct 2023
Contributor Spotlight: Mohammad Aflah Khanoriginal ↗
21 Sept 2023
The first New England RLHF Hackers Hackathonoriginal ↗
19 Sept 2023
EleutherAI's Thoughts on the EU AI Actoriginal ↗
26 Jul 2023
Minetester: A fully open RL environment built on Minetestoriginal ↗
8 Jul 2023
🐶Safetensors audited as really safe and becoming the defaultoriginal ↗
23 May 2023
Alignment Research @ EleutherAIoriginal ↗
3 May 2023
Transformer Math 101original ↗
17 Apr 2023
Exploratory Analysis of TRLX RLHF Transformers with TransformerLensoriginal ↗
2 Apr 2023
EleutherAI Second Retrospective: The long versionoriginal ↗
26 Mar 2023
The View from 30,000 Feet: Preface to the Second EleutherAI Retrospectiveoriginal ↗
2 Mar 2023

2022

Announcing GPT-NeoX-20Boriginal ↗
2 Feb 2022

2021

A Preliminary Exploration into Factored Cognition with Language Modelsoriginal ↗
25 Oct 2021
Multiple Choice Normalization in LM Evaluationoriginal ↗
11 Oct 2021
Downstream Evaluations of Rotary Position Embeddingsoriginal ↗
16 Aug 2021
What A Long, Strange Trip It's Been: EleutherAI One Year Retrospectiveoriginal ↗
8 Jul 2021
Why Release a Large Language Model?original ↗
2 Jun 2021
On the Sizes of OpenAI API Modelsoriginal ↗
24 May 2021
Evaluating Different Fewshot Description Prompts on GPT-3original ↗
24 May 2021
Finetuning Models on Downstream Tasksoriginal ↗
24 May 2021
Activation Function Ablationoriginal ↗
24 May 2021
Rotary Embeddings: A Relative Revolutionoriginal ↗
21 Apr 2021