"of" in "matter of" phrases
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. 19 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 reacts to the word "of" in set phrases like "a matter of time" and "a world of". It is negative on "of" in "the type of". Turning it down made the text get stuck repeating "of the ... of the".
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.
-
depression, and possibly post-traumatic stress disorder. He committed suicide on Sept. 29, 1998.⏎Bulling isn’t a game of “child’s play.” Suicide is becoming a more common result of teen bullying like
-
that strength training for muscle growth should use several different exercises per muscle group, the frequency of changing exercises is still a matter of debate. According to the latest recommendations, free-weight exercises involving complex movement patterns should
-
only be a matter of time before we got hold of a dead animal - because they are hunted - and
-
Divide⏎To die. I'm really scared that mom is going to cross the Great Divide any day now. The doctors are saying that it's only a matter of time.⏎divide by something⏎to perform mathematical division by a particular number. Can
-
it were just a matter of gathering new knowledge of distant planets and moons, human space travel would not be needed
-
Exploration Corp., a natural gas exploration company looking at ways to reduce its CO2 emissions.⏎Menova president Dave Gerwing says Trident knew it was only a matter of time before the federal government began imposing penalties on CO<|endoftext|>Multilateral Trade Negotiations
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: of the of the of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
also connecting to the gums and jawbone. If something happens to any of the layers of the tooth, the body can feel repercussions from the damage.⏎The type of issues an endodontist treats⏎A professional endodontist is a dentist who
-
through the water. A propeller determines the speed and the direction of the craft. A propeller contributes immensely to the functioning of the watercraft.⏎The type of the propeller is determined by the number of blades and the material that makes up
-
or deep? Explain the type of plate boundary it was near. Why are earthquakes common to this area?⏎Where was
-
to seek professional care from a specialist. A general dentist will not have the experience or tools to treat serious damage, where an endodontist does.⏎The type of issues an endodontist treats⏎While minor issues such as plaque buildup or even
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
- L16:1088 weakly excites 'of' after a noun phrase, as in 'of course'
- L18:821 weakly excites
- L7:2063 weakly suppresses
- L6:357 weakly suppresses
- L1:968 weakly suppresses
- L7:821 weakly excites
- L18:735 weakly excites
- L12:1020 weakly excites
- L13:2580 weakly suppresses
- L18:2364 weakly excites
- L6:368 weakly excites
- L17:317 weakly excites 'of' after a noun phrase
- L8:1955 weakly suppresses
- L5:363 weakly excites
- L17:1458 weakly suppresses
- L17:952 weakly suppresses
- L15:2775 weakly excites
- L14:836 weakly suppresses
- L14:849 weakly excites
- L9:359 weakly suppresses
- L17:527 weakly excites
- L2:2369 weakly excites
- L4:1489 weakly excites
- L10:3016 weakly suppresses
Feeds later layers
- L20:2599 suppresses
- L22:2882 suppresses
- L21:1401 weakly excites prepositions opening a phrase ('after', 'on')
- L20:1712 weakly suppresses
- L21:323 weakly suppresses
- L24:299 weakly suppresses
- L20:2093 weakly suppresses
- L20:679 weakly suppresses
- L26:711 weakly suppresses 'years' after an age number
- L21:732 weakly suppresses
- L23:258 weakly suppresses
- L20:337 weakly excites
- L23:355 weakly suppresses
- L20:318 weakly suppresses
- L25:2203 weakly suppresses
- L21:1609 weakly suppresses
- L22:322 weakly suppresses
- L20:2247 weakly suppresses
- L26:510 weakly suppresses
- L26:920 weakly suppresses
- L21:674 weakly suppresses comma after 'if you liked this article'
- L21:1018 weakly suppresses
- L20:353 weakly excites
- L22:811 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
- ematics
- ec
- ages
- inations
- 津
- ugged
- erek
- agination
- ugs
- enames
Suppresses
- 之旅
- bac
- submission
- viol
- ofs
- нибудь
- WithValue
- ña
- )[-
- _suite