"any" in negated sentences
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. 3 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 "any" when it follows a negation, such as "don't contain any", "isn't showing any" or "hasn't received any". Its negative side is weak and not clearly meaningful. Turning it up did not give a clear, on-topic effect in the steered text.
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.
-
or can be manufactured under circumstances that lack quality control.⏎It is known that more than 30 percent of the counterfeit drugs that are available today don’t contain any active ingredients whatsoever. “Counterfeit drugs include preparations without active ingredients, toxic preparations
-
retaliated.⏎- It is the heaviest fighting between the two bitter enemies since 2014 and isn’t showing any signs of slowing.⏎- An emergency meeting of the UNSC has been conducted for
-
. Let’s just say we wouldn’t go making any plans for distant space travel or a personal vegetable garden on the moon just yet.⏎
-
that oil would wash up on that island," he said. "I’m not aware that any significant oil did occur on that island.”⏎Newstead hasn’t received any reports, of areas crucial to bird nesting, that were damaged.⏎He attributes that
-
figure 182. Without exerting any pressure, carefully detach the lose threads form the signatures, and keep them aside.⏎
-
72 and they realized it could read the tapes and process the data. Allen had graduated and was enrolled at Washington State University. Since neither Gates nor Allen had any hardware design experience they were initially stumped. The computer community in Seattle at that
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 most important for you? I have been thinking about lately: what is the most important for you? I have been thinking about lately: what is the most important for you? I have been thinking about lately: what is the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
de Ribereños y⏎Armadores, an association of fishermen and producers in Mexico.⏎"Fishermen are desperate."⏎Gilly blames the most recent El Niño in the Pacific Ocean for⏎forcing some Humboldt squid to migrate away. Others
-
of strange attractors (J.C. Sprott). Attractors with dueling symmetry (C.A. Reiter). A new feature in Hénon's map (M. Michelitsch, O.E. Rössler
-
for the effects of age, sex, race, and alcohol consumption, depression was twice as frequent in those who had intermediate cardiovascular health and two and a half times more frequent in those who had poor cardiovascular health when compared to those who
-
essence that forms the basis of her aspect of manifestation; they are the essential content by which all worldly forms are shaped.⏎In Filianic Scripture, the Janyati are compared to the splitting of white light through a prism, creating a plethora
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
- L2:2253 excites
- L2:1891 weakly excites
- L2:1215 weakly excites
- L2:111 weakly suppresses
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L0:2681 weakly excites
- L2:3003 weakly suppresses
- L2:2328 weakly suppresses
- L0:1799 weakly suppresses
- L0:1224 weakly suppresses
- L2:1416 weakly suppresses
- L2:2806 weakly suppresses
- L1:2378 weakly excites
- L2:2696 weakly suppresses
- L1:146 weakly excites
- L0:2380 weakly excites
- L1:2683 weakly excites
- L2:510 weakly suppresses
- L1:1029 weakly excites
- L0:564 weakly suppresses
- L2:269 weakly excites
- L1:301 weakly excites
- L2:2173 weakly suppresses
Feeds later layers
- L4:731 excites
- L16:1009 excites
- L9:59 excites "a" meaning per (a day, a year)
- L23:2339 excites
- L17:1516 excites
- L16:1309 excites
- L17:2233 excites 'any' after negatives or 'more'
- L23:2403 excites unclear: 'no matter' and 'several' before reasons
- L17:2902 excites
- L4:137 weakly excites
- L5:9 weakly excites
- L25:283 weakly excites
- L16:2055 weakly excites
- L17:1524 weakly excites
- L17:2348 weakly excites
- L15:519 weakly excites
- L14:2546 weakly excites
- L4:1938 weakly suppresses
- L22:146 weakly excites
- L26:10 weakly excites
- L4:283 weakly excites
- L16:1910 weakly excites
- L14:1463 weakly suppresses
- L7:224 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
- nisi
- ")[
- 例外
- except
- […]
- 次要
- […]
- кроме
- amine
- [mid
Suppresses
- ycled
- ovie
- Din
- jeta
- glich
- ICLES
- edom
- filme
- :⏎⏎⏎⏎
- (||