'of' after a job 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. 11 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 'of' when it follows a job title, such as 'chief of', 'professor of' or 'director of'. Turning it up did not produce a clear theme in the written text, so its effect on the writing is hard to read.
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.
-
and Hispanic children have close to double the rate of obesity in comparison to white children in the U.S.,” says Dr. Elsie Taveras, chief of General Pediatrics at Mass General Hospital. “The behavioral risk factors during<|endoftext|>An Australian
-
"A landmark study on one of the most pressing problems facing society, balancing economic growth and ecological protection to achieve a sustainable future."⏎- Simon Levin, Moffett Professor of Biology, Department of Ecology and Evolution Behaviour, Princeton
-
not as simplistic as we've tended to imagine," says study co-author T.R. Kidder, PhD, professor and chair of anthropology in Arts and Sciences at Washington University in St. Louis.⏎"Our findings
-
protein modification. These tumor samples consistently tied to worse outcomes in children, suggesting a critical marker for predicting prognosis.⏎What's more, the researchers found that simple immunohistochemical staining methods in tumor biopsy samples can show whether the tumor has high or
-
of Music, University of Chicago⏎—Anthony Cascardi, University of California, Berkeley⏎—Jesse J. Prinz, John J. Rogers Professor of Philosophy, University of North Carolina, and author of Furnishing the Mind: Concepts
-
, including the number of the unique words and a list of the most frequent words, a Corpus Reader that lets you see all your text and select certain words and highlights
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 am a woman of many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many many
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
this question⏎Join a community of thousands of dedicated teachers and students.Join eNotes<|endoftext|>Is there something else you would like to know?⏎Use the alphabetical glossary listing to find a short definition to a medical or dental term. If you would like
-
services and evaluate video quality in six categories of digital defects on a scale of one to five (one being unwatchable and five being spectacular).”⏎“In addition, Comcast is working to establish local golden-eye programs
-
, but it also determines your blood type. There are two main blood groups, the ABO group and the Rh group. The ABO group is made up of A,
-
.⏎A thin layer of tissue is excised from<|endoftext|>In computer programming, a Either-OR is used as an OR operator to choose a value between 2 values. Either of Two words are often used to represent the truth value of a statement
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
- L10:2958 excites
- L10:1707 excites
- L7:1094 weakly suppresses
- L10:2012 weakly suppresses
- L9:1491 weakly suppresses
- L3:2478 weakly excites
- L10:1996 weakly excites
- L9:996 weakly excites
- L9:2257 weakly suppresses
- L8:2399 weakly suppresses
- L5:256 weakly excites
- L8:2573 weakly excites
- L4:978 weakly excites
- L10:2092 weakly suppresses
- L7:179 weakly suppresses
- L0:3047 weakly suppresses
- L2:1446 weakly suppresses
- L9:728 weakly suppresses
- L4:2504 weakly excites
- L7:2066 weakly excites unclear: technical word fragments, mostly engineering
- L8:2624 weakly suppresses
- L9:1416 weakly excites
- L10:2561 weakly suppresses
- L10:745 weakly suppresses
Feeds later layers
- L15:1163 excites
- L12:1311 suppresses
- L16:2938 suppresses
- L13:246 excites
- L15:2696 weakly excites
- L14:1670 weakly excites
- L12:301 weakly suppresses
- L16:1131 weakly suppresses
- L18:2108 weakly suppresses
- L12:1962 weakly excites
- L12:1794 weakly suppresses
- L15:1140 weakly excites
- L14:433 weakly excites
- L12:2464 weakly suppresses
- L12:1681 weakly excites
- L13:721 weakly suppresses
- L12:1085 weakly suppresses
- L18:2492 weakly suppresses
- L17:2681 weakly suppresses
- L12:2702 weakly excites
- L12:2019 weakly excites
- L23:2687 weakly suppresses
- L16:1863 weakly excites
- L14:2433 weakly suppresses
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
- essional
- @return
- Poster
- ieder
- ready
- 折扣
- зол
- getRequest
- icz
- 撞击
Suppresses
- аниз
- angement
- 芒
- eson
- 拉斯
- 芙
- ItemCount
- ас
- ocrin
- Ars