"the" before a named organisation or 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. 9 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 lights up on the word "the" when it starts a specific name, such as a journal, a society, a council or a convention. When it was turned up, the model's writing drifted into unrelated topics, and turned down it got stuck repeating "the the".
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.
-
provides information of a general nature and is designed for educational purposes only. If you have any concerns about your own health or the health of your child, you should always consult with a physician or other health care
-
in the herbal remedy plant St. John's wort, becomes super-toxic to viruses and cancer cells when exposed to light.⏎The results were published today in the Journal of the American Chemical Society by chemists from Iowa State University and the U
-
one side of an equilateral triangle and the circle touches the other two sides of the triangle. A smaller circle touches the larger circle and. . . .⏎X is a moveable point on the hyp
-
the US, this did not include the waste produced by the American owned maquiladoras in Mexico. It was really an act to ratify the Basel Convention on the control of transboundary movement of hazardous wastes and their disposal. Right now the issue
-
x=n.⏎SOURCE: Carpenter, Corbitt, Kepner, Lindquist, and Reys, 1981. Used by permission of the National Council of Teachers of Mathematics.⏎merical values and use it when x=
-
.<|endoftext|>Start a 10-Day Free Trial to Unlock the Full Review⏎Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share
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: what is the a question about I need to write a response Okay, so I'm thinking about something. Maybe that's a question. I need to write a response. I need to write a response Okay, I'm thinking
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
complication of liver cancer.⏎Generally, cirrhosis appears in at least 20% of patients within 20 years of infection. Cir<|endoftext|>Parts of the woods at Tangled Branches South are carpeted with blueberries
-
⏎- PRODUCING THEATER⏎- CREATING THEATER⏎- THE CREATIVE TEAM⏎- THE CREATIVE ARTISTS⏎- IMPLEMENTING THE DESIGNS⏎- PRODUCTIONG MANAGEMENT / STAGE MANAGEMENT / TECHN
-
composed of dark energy and dark matter.⏎Besides all of this, Rubin was an advocate for women in the sciences–especially as she was often the only woman in the field. For instance, she was the only astronomy major to graduate from the women
-
, Secondhand Pesticides. We found that many pesticides remained in the air near fields at high levels for hours, or even days after application. Anyone in the vicinity of such applications faced exposure.⏎Sept. 9
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
- L4:52 strongly excites
- L3:291 excites
- L6:640 suppresses "the" introducing a noun phrase after a name or preposition
- L1:59 excites 'the' after a verb or preposition
- L0:26 suppresses
- L8:759 excites
- L5:1172 excites
- L4:49 suppresses
- L6:426 suppresses
- L8:595 excites
- L6:61 suppresses
- L1:1796 excites
- L8:2257 suppresses
- L0:39 weakly excites 'the' right after 'On' in 'On the other hand'
- L6:443 weakly suppresses
- L4:163 weakly suppresses
- L1:2340 weakly suppresses
- L2:111 weakly suppresses
- L0:774 weakly suppresses
- L5:46 weakly excites
- L0:1905 weakly suppresses
- L3:59 weakly excites
- L2:1914 weakly excites
- L7:1716 weakly excites
Feeds later layers
- L17:30 strongly excites
- L16:792 strongly excites "the" inside a prepositional phrase
- L18:641 strongly excites
- L16:419 strongly excites
- L17:1481 excites 'the' after 'some of' or 'most of'
- L26:1941 excites
- L20:361 excites
- L18:799 excites
- L16:1776 excites 'the' after a verb or preposition
- L21:614 excites
- L12:435 excites
- L16:2062 excites "other" and contrast words
- L17:2970 excites
- L10:120 excites
- L18:1547 excites
- L17:2190 excites 'the' starting an object phrase
- L19:427 excites
- L20:2374 excites "the" after verbs like laid, reaching, raises
- L19:1376 excites
- L20:1246 excites
- L11:875 excites
- L16:2266 excites 'the' before study or journal names
- L11:1168 suppresses
- L25:2373 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
- same
- aforementioned
- same
- oret
- latter
- basics
- following
- latest
- association
- respective
Suppresses
- 很喜欢
- 毌
- icult
- uously
- ⏎
- entially
- antas
- `=
- 累了
- 因地制宜