"of" in "some kind of" or "some type of"
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. 1 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 phrases like "some kind of" or "some type of". It is quiet, or slightly negative, on words like "or" in "either-or" choices. Turning it up or down gave text that wandered off in different directions, so it does not clearly change the topic.
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.
-
You may need to invest in some type of pesticide;there are many DIY chemical-free recipes available nowadays too.⏎As the summer
-
to do some type of exercise every day. It is best to do some kind of aerobic activity without stopping for at least 20 to 30 minutes each
-
to yourself. For example, thinking that everything people do or say is some kind of reaction to you. You also compare yourself to others, trying to determine who’s
-
embassy. They are often set up between nations that do not have official diplomatic relations. They offer a way for two nations that have tense relations to continue some kind of working contact. The U.S. has operated “interest sections” in places like
-
and also points to separate research which estimates that up to 22% of crashes could be caused by some kind of distraction; and drivers who perform a secondary task at the wheel, like using a
-
a few days, the CNT and the Socialist General Confederation of Workers (UGT), which since December 1933 had been discussing some kind of working-class alliance, planned a national general strike. It began on October 4
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 am thinking. So I am thinking. I am thinking. So I am thinking. So I am thinking. So I am thinking. So I am thinking. So I am thinking. So I am thinking. So I am thinking.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of motor oil, there are three pieces of information: the API service rating, the viscosity grade and the “Energy Conserving” indicator (the oil either is or it isn’t).⏎API service rating⏎This
-
Piper urges us to think for the glory of God. He demonstrates from Scripture that glorifying God with our minds and hearts is not either-or, but both-and. Thinking carefully about God fuels passion and affections for God. Likewise, Christ
-
another word this?⏎- Which is short form?⏎- Which is or that is?⏎- What is the means of 7?⏎- What Why means
-
programming language.⏎Python does not have a static checking phase,⏎The (C)Python *interpreter* doesn't. Other Python implementations (existing or hypothetical) may or may not have a static checking phase.⏎But static tools need specifications beyond (ie
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
- L0:2651 weakly suppresses
- L0:1812 weakly suppresses
- L0:2793 weakly suppresses
- L0:2537 weakly suppresses
- L0:1945 weakly suppresses
- L0:146 weakly suppresses
- L0:395 weakly excites
- L0:1433 weakly suppresses
- L0:373 weakly excites
- L0:1100 weakly excites
- L0:2616 weakly suppresses
- L0:376 weakly excites
- L0:1394 weakly suppresses
- L0:983 weakly suppresses "similar" after "are"
- L0:1475 weakly suppresses
- L0:78 weakly excites the 'S' in 'U.S.'
- L0:1720 weakly excites
- L0:942 weakly excites
- L0:895 weakly excites
- L0:5 weakly excites
- L0:189 weakly suppresses
- L0:427 weakly suppresses
- L0:260 weakly excites
- L0:2817 weakly suppresses
Feeds later layers
- L7:2170 weakly suppresses
- L14:2276 weakly excites opening bracket after a term or number
- L10:2235 weakly excites
- L7:712 weakly excites
- L19:1835 weakly suppresses
- L6:876 weakly excites
- L16:51 weakly suppresses
- L12:1823 weakly excites
- L3:1114 weakly suppresses
- L17:986 weakly excites
- L12:2662 weakly suppresses
- L18:1975 weakly suppresses
- L16:184 weakly suppresses
- L3:710 weakly excites
- L11:1667 weakly excites
- L14:346 weakly excites
- L5:1691 weakly suppresses
- L13:1592 weakly excites
- L21:1993 weakly excites
- L16:47 weakly suppresses
- L16:1402 weakly suppresses
- L14:1296 weakly excites
- L11:2954 weakly suppresses
- L16:454 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
- 老师的
- мент
- lication
- 師
- 师
- ulses
- case
- ilies
- isted
- 생
Suppresses
- 成熟
- ja
- eno
- Frem
- frem
- buiten
- décou
- 宰
- compte
- Mature