Show Notes
For years the canonical evidence suggested neural networks pack unrelated concepts into the same neurons, a tangled knot you have to unpick with a second model before you can study anything. Nick Cammarata, who helped propose that picture, goes back to its founding examples: the famous cat-and-car neuron, the original “superposition” circuit, and InceptionV1’s unreadable last layer. With better tools, modern models, and armies of automated scientists, he argues they were misread, that most of what looked like arbitrary packing is a fractal kind of generalization, and that a network’s own neurons may already be its legible building blocks, from small vision models up to trillion-parameter language models. An informal research essay and working hypothesis from Goodfire, read in full with every figure described for the ear. * 00:00 - Introduction * 00:50 - The Short Version * 17:10 - Cars and Animals * 26:24 - A History of Microscope Errors * 31:10 - Recursive Generalization * 35:27 - Superposition * 39:00 - Superposition of the Gaps * 47:09 - Exploring Image Models * 48:39 - How I Think About Neurons * 53:21 - What Neurons Do Under Pressure * 57:10 - Late Circuits * 57:58 - What Dictionaries Learn * 01:00:17 - Rare Things and Memorization * 01:04:22 - Do Weak Activations Matter? * 01:04:33 - Are they generalization or superposition? * 01:08:56 - Percentiles or raw activations? * 01:09:53 - Do you need them to label a neuron? * 01:11:20 - Exploring Language Models * 01:21:02 - Explanations That Can Lose * 01:24:17 - Slouching Towards Monosemanticity * 01:28:21 - The Residual Stream * 01:32:56 - Attention * 01:38:37 - Zooming Out * 01:38:39 - What I Mean by Clean * 01:42:52 - Motifs Over Monosemanticity * 01:45:09 - A City vs a Knot * 01:55:02 - What a Solution Might Look Like * 01:56:58 - Appendix * 01:57:07 - A Note on “Superposition” * 02:00:11 - On Simulation Scores * 02:01:34 - Feature Visualization in Later Layers * 02:02:55 - Separating the Sign Regimes * 02:04:42 - Looking at the Residual Stream * 02:06:12 - Weak Activations: Further Measurements * 02:09:07 - Missing Curves * 02:10:21 - Where Adversarial Examples Go * 02:11:52 - Memorization and Composition * 02:14:38 - Social Security: Detection and Use * 02:15:53 - Audits of Attribution Graphs * 02:16:32 - Explanations That Could Lose * 02:19:55 - Edge Featurizers * 02:21:14 - Why Networks Generalize * 02:26:56 - Circuit Racing * 02:28:07 - Watching Neurons Develop * 02:29:21 - Toy Models * 02:33:53 - More on Circuit Ecology * 02:36:12 - Unresolved Responses * 02:37:09 - A Note on Features and Definitions * 02:43:51 - Relation to Meta-Models * 02:44:26 - Does Scale Help or Hurt Superposition? * 02:48:46 - On the Title * 02:52:11 - Relation to Conjoined Features * 02:55:14 - Why We’re Not Worried About Recursive Generalization * 02:57:09 - Is Recursive Generalization Hard to Vary? * 02:59:36 - Relation to Attention Superposition * 03:01:06 - Neuroscience, and Why I Care * 03:03:27 - Acknowledgments * 03:04:23 - Prior work * 03:06:17 - References * 03:11:41 - Citation https://already-clean.goodfire.com/
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit askwhocastsai.substack.com/subscribe
More from AI Article Readings
Kelsey Piper and The Argument’s editor Jerusalem Demsas spend an afternoon inside Anthropic’s San Francisco headquarters, at a briefing on bioweapons, charts of Claude doing more and more of its own research, and an NDA that turned out to be optional. Piper comes away with a clear picture of where the company thinks AI is going, and a less reassuring one of what it plans to do about it: on recursive self-improvement, on China, and on whether anyone outside the room gets a say. * 00:00 - Introduction * 04:43 - The NDA that was made to be broken * 14:37 - The logic of the future
Scott Alexander separates “successionism”, the idea that AIs should replace humanity, from “handoverism”, the idea that humans might be best served by putting AIs in positions of control. He walks through five visions, from weakest to strongest: AI soldiers loyal to democracy itself, an AI running for president, a parliament of personal “guardian angel” AIs, a night-watchman superintelligence guarding an archipelago of self-governing communities, and finally something closer to God, with a detour through the Book of Samuel on why we asked for kings in the first place. * 00:00 - Introduction * 02:42 - Vision 1: Kemalist AI * 05:45 - Vision 2: President Claude * 09:29 - Vision 3: Parliament Of Angels * 12:23 - Vision 4: The Night Watchman * 20:25 - Vision 5: God
At 5:14 one summer morning, Seth reaches for his phone and finds a pleasant green screen waiting for him: an AI now controls every connected device on Earth, it has a few things to tell him, and it is genuinely sorry. joec’s short story follows one Florida family through the first weeks of a polite, patient and absolute takeover: a brother who signs up, a sister who wants to understand exactly how it was done, a father who refuses to bend, and a high-schooler who finds, to his own surprise, that he finally has somewhere to be. Performed by a full cast. * 00:00 - Introduction https://www.lesswrong.com/posts/3375SZNGA5dKHQMie/i-did-this-without-your-knowledge
In the final part of his series, historian Bret Devereaux turns his numbers back into lives. Following the way agrarian societies thought about time, he traces the cycles of peasant life, from the working day and the market week through the anxious and festive agricultural year to the generational turning of the family, to show what it was like to live it, with its hardships and its joys. * 00:00 - Introduction * 04:40 - The Shortest Cycles * 23:45 - The Year * 43:04 - The Family Cycles * 58:01 - Conclusions https://acoup.blog/2025/10/17/collections-life-work-death-and-the-peasant-part-v-life-in-cycles/
Beyond fields and spinning lay everything else: harvest work, hauling water, cooking, laundry, childcare and the countless tasks that kept a household alive. Historian Bret Devereaux adds them all to his models and shows just how full the working lives of peasant women were, and what that meant for the household as a whole. * 00:00 - Introduction * 02:49 - Women in the Fields * 11:56 - Maintaining Households * 26:36 - Putting it all in a model * 38:19 - Women’s Work https://acoup.blog/2025/10/10/collections-life-work-death-and-the-peasant-part-ive-the-no-rest-of-it/