Top all-time · #4261
Goodfire
Company blog · goodfire.ai/blogaiscience
Goodfire is the company blog at goodfire.ai, writing about ai and science. Since 2024, 17 different Hacker News readers have linked it 19 times, most often “The World Inside Neural Networks”. No feed was found, so read it on the site itself.
Too few sampled comments (6) to say how often HN recommends it.?
Why HN links it
HN links Goodfire's pages as sources on LLM interpretability: sparse autoencoders for Llama 3.3 70B, latent-space manipulation, and research on models detecting evaluation. Comments cite them without judging the writing.
Summarised by Claude from a sample of 6 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Goodfire is an AI interpretability research lab focused on understanding and intentionally designing advanced AI systems.
Most-linked posts?
- The World Inside Neural Networks2 people · 2 mentions · first linked 2026-05-07
- A Geometric Calculator Inside a Neural Network2 people · 2 mentions · first linked 2026-05-14
- Show HN: Llama 3.3 70B Sparse Autoencoders with API access2 people · 2 mentions · first linked 2024-12-23
What Hacker News says?
We're planning to release the weights once we do a moderation pass. Our SAE was trained on LMSys (you can see this in our accompanying post: https://www.goodfire.ai/papers/mapping-latent-spaces-llama/ ). No images in training - 3.3 70B is a text-only model so it wouldn't have made sense. We're exploring other modalities currently though. SAE is a basic ReLU one. This might seem a little backwards, but I've been conc…
Comment linking /papers/mapping-latent-spaces-llama/
on “Show HN: Llama 3.3 70B Sparse Autoencoders with API access”Submitted Show HN: Llama 3.3 70B Sparse Autoencoders with API access as a story · 201 points
HN discussionSee also the work being done by GoodFire AI: https://www.goodfire.ai/ They now have an API that allows for dynamic exploration and manipulation of the latent space for LLama 8-70B models (think Golden Gate Claude). They also open sourced the sparse auto-encoders that (in part) allow for this: https://huggingface.co/Goodfire/Llama-3.3-70B-Instruct-SAE-l...
Comment linking /
on “Transformer^2: Self-Adaptive LLMs”Submitted What if you could look inside your AI? Understanding and steering Llama 3 as a story · 21 points
HN discussionYou literally can do a kind of model PCA, using the Hessian (matrix of second derivatives of the loss function w/r/t the parameters, aka the local curvature of the loss landscape), and diagonalizing. These eigenvectors and eigenvalues (the spectrum of the Hessian) tend to be power-law distributed in just about every deep NN you can think of [1]. That is, there are a few "really important" (highly curved) dimensions…
Comment linking /research/understanding-memorization-via-loss-curvature
on “Billion-Parameter Theories”Submitted The World Inside Neural Networks as a story · 3 points
HN discussionThe fact that models can tell if they are being evaluated has been established by multiple research teams outside of the core AI vendors themselves. - https://metr.org/evaluations/gpt-5-report/ - "These behaviors included demonstrating situational awareness within its reasoning traces, sometimes even correctly identifying that it was being evaluated by METR specifically" - https://www.goodfire.ai/research/verbalized…
Comment linking /research/verbalized-eval-awareness-inflates-measured-safety
on “OpenAI's accidental cyberattack against Hugging Face is science fiction”Submitted Neural Geometry in Vision Models with Block-Sparse Featurizers as a story · 2 points
HN discussionWhy can't a next token predictor do math? Humans aren't calculators either, but we can do math. If you want proof just look at the benchmarks. Modern frontier models can get basically perfect accuracy on American Invitational Mathematics Examination tests: https://matharena.ai/?comp=aime--aime_2026 If you want an explanation of how they do math, we've found geometric calculators inside their neural networks: https:/…
Comment linking /research/a-geometric-calculator
on “LLMs Will Replace 8-Track Duplication Engineers”Submitted The World Inside Neural Networks as a story · 2 points
HN discussionIs this in any way similar to Goodfire's work? https://www.goodfire.ai/research/rlfr#
Comment linking /research/rlfr
on “Show HN: Cactus Hybrid: We taught Gemma 4 to know when it's wrong”Submitted Using Interpretability to Identify a Novel Class of Alzheimer's Biomarkers as a story · 2 points
HN discussionSubmitted Replicating Circuit Tracing for a Simple Known Mechanism as a story · 2 points
HN discussionSubmitted Predictive Data Debugging: Reveal and Shape What Models Learn Before You Train as a story · 1 points
HN discussionSubmitted Can SAEs Capture Neural Geometry? as a story · 1 points
HN discussionSubmitted A Geometric Calculator Inside a Neural Network as a story · 1 points
HN discussionSubmitted A Geometric Calculator Inside a Neural Network as a story · 1 points
HN discussionSubmitted Intentionally Designing the Future of AI as a story · 1 points
HN discussionSubmitted Under the Hood of a Reasoning Model as a story · 1 points
HN discussion
