"a" after a name, introducing a title
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 1 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the word "a" right after a person's name in a news quote, as in "Natalia Trayanova, a professor". It is the spot where the writer is about to describe who that person is. Turning it up did not produce a visible change in topic.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
Department of Biomedical Engineering at The Johns Hopkins University Schools of Engineering and Medicine and the project’s principal investigator… Continue reading.⏎Natalia Trayanova, a professor
-
vehicles’ directions and velocities to make decisions.⏎“The motivation is, ‘What can we do with as little information as possible?’” says Alyssa Pierson, a postdoc at CSAIL and first author on the new paper. “How can
-
anxiety disorder,” he said⏎“It’s interesting work that fits nicely with recent work suggesting that testosterone influences social motivation and perception,” said Pranjal Mehta, a psychologist at Columbia University who was not involved in the study. Future studies should test
-
campaign on energy policy. “We desperately need to find a new form of energy.”⏎Colin Campbell, a retired geologist, says governments are too caught up in short-term issues to focus
-
Lab. Located on the third floor of the Student Learning Centre, where the DME helps students incorporate these technologies into class projects.⏎Kieran Ramnarine, a DME student advisor who helped Luksatrio with his project, said that
-
, most people have focused on larger systems that produce 1 megawatt of power or more and can replace traditional power plants," said Vincent Sprenkle, a co-author on the paper and chief engineer of PNNL's solid oxide fuel
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: "I think the problem is that the problem is not a problem." What do I mean by that? What do I mean by that? What do I mean by that? What do I mean by that? What do I mean by that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
cycle and biology of Xyleborus glabratus.⏎Adult females construct galleries in the sapwood and inoculate the galleries with a fungus (Ophiostoma sp., vascular wilt pathogen). The adults and larvae feed on fungi and not
-
the yeast-like fungus Candida albicans or occasionally by other species of Candida. Vulvovaginitis due to Candida presents with an erythema of the vaginal mucous membrane and the vulval skin, itching
-
of mammals act as natural reservoirs of Trypanosoma cruzi (the causal agent of Chagas disease) across the American continent.
-
or vesicles and rupture, leaving a peeling edge. Involvement of groins sometimes coexists. Perianal candidiasis presents with erythema, soreness and irritation, and subsequent spread along the natal cleft is
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L0:1224 suppresses
- L0:1301 suppresses
- L0:178 suppresses
- L0:1042 weakly excites "This" starting a labelled sentence
- L0:300 weakly excites
- L0:351 weakly suppresses
- L0:1408 weakly excites
- L0:1065 weakly excites
- L0:2118 weakly excites
- L0:2799 weakly suppresses
- L0:1299 weakly suppresses
- L0:2259 weakly suppresses
- L0:955 weakly suppresses
- L0:909 weakly suppresses
- L0:1645 weakly suppresses
- L0:313 weakly suppresses "and" joining paired nouns
- L0:2280 weakly suppresses
- L0:2743 weakly excites
- L0:836 weakly suppresses
- L0:2572 weakly suppresses
- L0:970 weakly suppresses
- L0:1025 weakly excites
- L0:652 weakly excites
- L0:2197 weakly excites
Feeds later layers
- L11:520 weakly excites
- L16:1735 weakly excites
- L21:360 weakly suppresses 'such' introducing examples
- L3:402 weakly excites
- L4:2267 weakly suppresses
- L20:1733 weakly suppresses
- L3:1938 weakly excites
- L14:1776 weakly suppresses
- L3:355 weakly excites
- L10:2952 weakly excites
- L26:804 weakly excites
- L7:22 weakly suppresses
- L20:1127 weakly excites
- L3:799 weakly suppresses
- L12:2456 weakly excites
- L14:2700 weakly suppresses
- L10:2026 weakly excites
- L13:1613 weakly excites
- L23:691 weakly suppresses 'it' before 'is important to'
- L9:2146 weakly suppresses
- L10:1124 weakly excites
- L14:819 weakly suppresses
- L2:2642 weakly suppresses
- L21:359 weakly excites
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- .appspot
- ��
- SSERT
- GetInstance
- =edge
- (;;
- bergen
- 梦见
- UPPORT
- Anthem
Suppresses
- iqu
- ten
- 几张
- istr
- exas
- ivas
- tar
- tu
- atcher
- 不留