"not" in "if not" contrasts
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. 10 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 "not" (or "less") in set phrases such as "if not most," "if not being taught here" or "much less resolved," where a second, contrasting idea is being added. The negative side is mostly odd punctuation in citations and links. Turning it up made the output repetitive and unrelated.
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.
-
four or five in their regular government schools. Despite their age, they still don’t know basic math. The toddlers are their siblings, who the girls — if not being taught here — would be taking care of at home, while
-
enzel found that restorative but not retributive responses led to greater forgiveness. Taken together, Wenzel’s studies suggest that the offender’s behavior and not his punishment will become the focus.⏎That’s not to say that there’s no
-
The center of the retina is called the macula.⏎This is the most sensitive area of the retina used for reading vision and straight-ahead vision. Retinal and specifically macular diseases can cause clouding and distortion of vision, loss of peripheral vision
-
member states.⏎The rapid growth of the socio-cultural complexities of society, also triggered by the fast development of scientific thought in the XXI century, in general, and the Internet, in particular, entails a sharp increase of the social diversity. This
-
nutritional and / or pharmacological.⏎It is very important to note that in addition to external factors
-
the entrepreneurial subject while encouraging a value system that promotes self-interest, if not an unchecked selfishness.1 Since the 1970s, neoliberal
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 first thing to note is that the first thing to do is to realize that if something is... in the context of time. The question is that is, the first thing to do is to realize that the first question is to understand.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Vitamin C and Immune Function, 2017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%2
-
-53. http://dx.doi.org/10.1080/13527266.2011.620764⏎BELK, R. (201
-
2017)⏎|Council of Trent|⏎|Accepted by||Catholic Church|⏎|Fifth Council of the Lateran (1512–1517)|⏎|First Vatican Council (1869–
-
oietin in the brain improves exercise performance in mice without inducing erythropoiesis. FASEB J. doi:10.1096<|endoftext|>Valentine’s Day Task Cards (2nd – 4th Grade)⏎
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
- L9:880 suppresses
- L8:2127 excites
- L7:1156 suppresses
- L9:2892 suppresses
- L8:1680 weakly suppresses
- L8:803 weakly suppresses
- L7:2107 weakly excites
- L7:1846 weakly suppresses
- L5:1924 weakly suppresses
- L1:2725 weakly suppresses
- L8:2839 weakly excites
- L2:2302 weakly suppresses
- L1:2093 weakly suppresses
- L8:1023 weakly excites
- L2:1537 weakly excites
- L8:2617 weakly suppresses
- L8:911 weakly suppresses
- L5:778 weakly suppresses
- L9:2936 weakly suppresses
- L1:2664 weakly suppresses
- L7:2906 weakly suppresses
- L3:1212 weakly suppresses
- L1:1825 weakly excites
- L7:498 weakly suppresses
Feeds later layers
- L16:2578 excites
- L14:1102 weakly excites
- L20:1707 weakly excites
- L13:2356 weakly excites
- L15:1699 weakly excites
- L14:1763 weakly excites
- L24:1091 weakly excites
- L13:1757 weakly excites
- L18:2305 weakly suppresses
- L12:1443 weakly excites opening parenthesis after a name or term
- L26:2098 weakly suppresses
- L25:1341 weakly suppresses
- L16:2746 weakly suppresses
- L14:853 weakly excites
- L12:1034 weakly excites
- L15:752 weakly suppresses
- L14:651 weakly excites
- L14:2308 weakly excites
- L16:2719 weakly suppresses
- L14:268 weakly excites
- L15:3004 weakly excites
- L15:1474 weakly excites
- L14:439 weakly excites
- L14:1963 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
- まり
- 会被
- ecessarily
- its
- Periph
- igidBody
- malink
- /releases
- OrDefault
- forced
Suppresses
- �建
- �数
- uja
- бит
- CATEGORY
- 她是
- オー
- ´t
- 蒽
- ships