comma before a clause that contrasts a claim
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. 21 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 commas and quote marks that come right after a claim, where the sentence is about to add a contrast or an explanation. When turned up, the model got stuck repeating the word 'when'. Turned down, it repeated 'though'.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
in the concentration of species in solution.⏎It’s frustrating that the article makes it seem like the applications of a really awesome discovery in pure science are mere moments away, rather than explaining why a discovery like this is awesome and encouraging peoples’ interest in
-
species as a group often has been exaggerated and misrepresented. For example, ecologists and conservationists often describe non-native species as the world’s second greatest extinction threat, despite the fact that existing data shows this clearly not to be the case. It
-
tendency, in some quarters, to jump to a solution like bariatric surgery, and rhapsodize about what an acceptable safe effective treatment alternative it is, without ever examining the underlying reasons for the obesity.⏎Reasons for obesity⏎Probably
-
tips for self-care for a healthy, active life.⏎1. Don’t lose weight⏎Many people spend a lifetime trying to restrict their eating to be ‘thin’, and it’s no wonder when this is so often portrayed as the height of fashion
-
probably noticed that many executives – and customers – don’t understand the difference between these terms. They most likely go as far as referring to them as the same processes, which they’re definitely not. Let’s figure out
-
and not serious.⏎Many people think they are constipated when, in fact, their bowel movements are regular. For example, some people believe they are constipated, or irregular, if they do not have a bowel movement every day. However,
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: though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though though
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
---⏎Search for questions⏎Still looking for answers? Try searching for what you seek or ask your own question.⏎Posted 31 Aug 2009 • 1 answer⏎Posted 20 Jan 2010 •
-
most common. In countries where only monogamous marriage is legally valid, de facto polygamy is generally permitted, as long as adultery is not illegal. In areas like these, where polyg
-
without any cross-task interference.⏎Its modular nature affords extensive customisation. The major functional components such as the number of threads, caches and DSP resources are optional, giving the SoC designer the opportunity to influence performance, power and die size to
-
used depends on the condition being treated. Generally, most practitioners recommend 5 to 15 grams of dried chrysanthemum, usually in an infusion or decoction. Some practitioners recommend a type of chrys
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
- L14:35 suppresses
- L11:223 suppresses
- L18:366 excites comma inside numbers and dates
- L13:1219 weakly suppresses
- L20:750 weakly suppresses unclear: chinese academic text, 'when' loops
- L12:158 weakly excites
- L20:380 weakly excites
- L7:471 weakly suppresses
- L15:184 weakly suppresses
- L9:1128 weakly suppresses
- L5:1465 weakly suppresses
- L2:740 weakly excites
- L11:2745 weakly suppresses
- L14:2525 weakly excites
- L12:325 weakly suppresses
- L16:441 weakly excites
- L10:556 weakly suppresses
- L18:409 weakly suppresses
- L17:118 weakly excites comma after a clause ending
- L19:1983 weakly excites
- L17:2541 weakly suppresses comma before but also in not only constructions
- L6:2407 weakly suppresses
- L20:49 weakly suppresses
- L18:646 weakly excites
Feeds later layers
- L25:633 strongly excites
- L23:1645 excites
- L27:659 excites
- L23:449 excites
- L22:915 weakly suppresses digits inside historical years
- L23:677 weakly suppresses
- L27:1419 weakly excites
- L24:876 weakly suppresses
- L23:728 weakly suppresses
- L26:1199 weakly suppresses
- L25:1997 weakly excites
- L26:388 weakly suppresses
- L24:3055 weakly excites
- L27:1084 weakly suppresses
- L27:2352 weakly suppresses
- L22:2604 weakly suppresses
- L25:1593 weakly excites
- L23:2814 weakly excites
- L23:793 weakly excites
- L22:636 weakly excites
- L23:1157 weakly excites
- L26:850 weakly excites
- L22:1958 weakly suppresses
- L27:2168 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
- when
- when
- 时
- 的时候
- When
- When
- cuando
- 当
- →when
- quando
Suppresses
- though
- though
- although
- Though
- Though
- 一回事
- although
- aunque
- Although
- 虽然