'and' joining paired adverbs
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
The neuron fires on the word 'and' when it joins two adverbs or action words, like 'easily and quickly' or 'freely and uncontrollably'. It is negative on the '-ly' ending of an adverb. Turning it up gave repetitive or off-topic text, so that result does not help explain it.
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.
-
eyes and ears” that check food from other countries to make sure they’re safe.⏎- Improving the traceability of foods so that their origins can be easily and quickly traced and identified in the event of a recall.⏎- Allowing the FDA
-
personal vow never to be like their parent.⏎The venter style is at the opposite end of the anger management continuum. Unlike the suppressor, this person freely and uncontrollably expresses their anger. They take no prisoners and have no mercy in
-
in infamy, the United States of America was suddenly and deliberately attacked by naval and air forces of the Empire of Japan,” said President Franklin
-
of view, no more than 10 meters away. Even in the dark, I immediately recognized the shape of a large female brown bear. As I slowly and quietly backed away, two cubs ran up the bank and onto the road where
-
skills without any emphasis on creating a piece of art. Children are, by nature, process oriented. Adults are, by nature, product oriented. Adults can easily and unknowingly shift the focus of art activities to creating an end product. As adults
-
383 stated:⏎"Except as otherwise required by statute, matters of official record shall be made available, in accordance with published rule, to persons properly and directly concerned, except information held confidential for
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: I want to be a teacher in a small town. I have a lot of students and I think the town is small. I have a 1200-1800. I think the best teacher in the country,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, laced with methane and various hydrocarbons – and since the arrival of the Cassini spacecraft in July 2004 it has only grudgingly surrendered its secrets.⏎Titan’s northern hemisphere was in the dead of winter when Cass
-
Alzheimer's disease, Parkinson's disease, diabetes, heart disease, multiple sclerosis, Lou Gehrig's disease, severe burns, and many more illnesses and injuries, literally allowing the body to heal itself. These amazing cells can come from many sources besides
-
whole poem, but we're just getting into it now, thirty-nine lines in. Lowell illustrates how hard, and how steadfastly, the soldiers fought for one another, and for the Republic.<|endoftext|>By the
-
04:30 AM⏎Capturing carbon dioxide can be done in several ways, but the most unusual approach by far is to literally feed the greenhouse gas to CO2-hungry algae.⏎Several companies have attempted
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
- L8:725 strongly suppresses
- L8:2112 suppresses
- L8:2991 suppresses
- L7:969 excites
- L6:2329 excites
- L8:2822 excites
- L6:1684 suppresses
- L5:1327 suppresses
- L5:2281 weakly suppresses
- L4:2704 weakly excites the word 'systematic' in academic text
- L5:2349 weakly excites
- L8:2961 weakly excites
- L4:521 weakly suppresses
- L3:950 weakly excites
- L7:320 weakly excites
- L7:1186 weakly excites
- L6:2746 weakly excites
- L6:1885 weakly excites
- L8:1101 weakly excites
- L5:3010 weakly excites
- L6:2158 weakly suppresses
- L8:1350 weakly suppresses
- L4:1711 weakly suppresses
- L5:2597 weakly suppresses
Feeds later layers
- L12:1295 excites
- L20:1126 excites
- L16:1819 excites
- L20:2018 excites
- L12:431 excites
- L15:2736 excites
- L11:1210 excites
- L10:1335 excites
- L26:391 excites
- L10:1756 excites
- L10:991 excites
- L12:2773 excites
- L13:1424 excites
- L13:1792 excites
- L17:2481 excites
- L10:346 excites "each" in "each other" phrases
- L21:1594 weakly excites
- L19:2838 weakly excites
- L27:684 weakly excites
- L16:1577 weakly excites
- L19:2415 weakly excites
- L16:1374 weakly excites
- L12:2419 weakly excites
- L20:762 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
- ually
- edly
- uellement
- urally
- ingly
- astically
- ically
- ически
- mente
- ously
Suppresses
- ainting
- arkers
- aña
- LOGY
- annis
- berman
- searchData
- []{"
- curities
- agna