'other' after 'among' in lists
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 lights up on the word 'other' in phrases like 'among other things', and on similar words such as 'many' in 'among many other applications'. When it was turned up, the model's writing drifted into repeated, unrelated phrases.
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.
-
, cannabis is produced from cannabis leaves, and marijuana, in turn, is a plant that is grown, among other things, for consumption and to make users ‘high’. Thus, we can safely
-
Hugo Chávez was elected for a third six-year term.Top of page⏎Venezuela had one of the highest murder rates in Latin America due, among other factors, to the uncontrolled availability of firearms and ammunition. There also were concerns
-
their study had found correlation rather than causation between nuclear and renewables, and emissions, they based their conclusions on the assumption of causality, he said, among other things.⏎Interpreting the correlation between events or trends as causality is a
-
pressure sensors. These electronics can one day be wrapped around prosthetic hands to provide a sense of touch or to create wearable electronics for long term body sensing, among many other possible applications. The team’s devices are so sensitive that they can detect the
-
dropower and invented, among many other things, a bladeless steam turbine, the first remote-controlled torpedoes and
-
were allowed to work and shop there but could not live there. Application of the Group Areas Act became very lax in the 1980s, among other things because courts were not able to handle all the cases, and when the Act
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: the number of the last two digits, the number of the year, and the month/day, and the day of the week... the number of the last two digits, in some cases, in some cases, the day of the week,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to cooler areas. Keeping this in mind, the refrigeration cycle consists of seven stages:⏎1. Compression of the hot gas;⏎6. Evaporation; and⏎A basic vapor-compression refrigeration system consists of four primary components: a meter
-
Keywords: viral marketing, word-of-mouth marketing, consumption patterns, information traits, credibility.⏎ANTHEUNIS, M.L., & REIJMERSDAL, E.A. (2012). Social
-
addus VC, Light RW. Pleural effusion. In: Mason RJ, Murray JF, Broaddus VC, et al., eds. Murray and Nadel's Textbook of Respiratory Medicine. 5th ed.
-
page⏎Patterns of tree-liana interactions: distribution and host preference of lianas in a tropical dry evergreen forest in India.⏎Vivek, P., & Parthasarathy, N.⏎Lianas are structurally dependent life-form
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
- L10:74 strongly excites
- L9:92 suppresses "and" or "or" joining two items
- L8:1605 suppresses
- L10:667 excites commas inside lists of names
- L8:1783 excites
- L5:548 excites
- L10:2717 excites
- L5:776 excites
- L6:620 excites "and" in lists of three or more
- L8:986 excites
- L8:664 excites
- L8:1789 excites
- L5:531 excites
- L4:57 suppresses
- L6:363 excites
- L10:3020 weakly suppresses
- L8:1046 weakly suppresses
- L8:1047 weakly excites
- L8:1238 weakly excites
- L9:1656 weakly excites
- L6:41 weakly excites
- L5:965 weakly excites
- L1:2591 weakly suppresses
- L5:880 weakly excites
Feeds later layers
- L12:1440 strongly suppresses
- L16:2442 strongly suppresses
- L14:1116 suppresses 'and' inside lists
- L17:919 excites
- L14:2054 excites
- L14:1635 excites end of document token
- L13:1941 suppresses
- L15:132 suppresses
- L15:3020 suppresses
- L15:2277 suppresses
- L18:486 suppresses
- L15:60 suppresses
- L13:497 suppresses
- L13:1734 suppresses sentence end right before a line break
- L12:1603 excites
- L13:1389 excites
- L14:449 weakly excites
- L16:1447 weakly excites
- L12:925 weakly suppresses
- L15:273 weakly suppresses
- L12:1884 weakly excites
- L13:1063 weakly excites
- L14:2531 weakly suppresses
- L14:683 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
- נת
- MPG
- 好不好
- PACE
- LIK
- PATCH
- 皇后
- descricao
- 吠
- besch
Suppresses
- iments
- uml
- elist
- ctions
- ibel
- 两项
- iem
- erring
- ành
- oot