"to" before an infinitive verb
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. 6 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 "to" when it comes right after a word like "important," "successful" or "took it upon himself" and is about to introduce an action, as in "important to make sure." When it was turned up, the model's replies did not show a clear theme.
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.
-
irrigation through early plant development (several weeks), will be necessary in our dry summer climate. In shad areas, like the one above, its more successful to plant out roots stalk instead of seeding, but I always throw something down, just
-
.”⏎Since the IT band’s primary function is to stabilize the pelvis, thigh, and knee during movement, its important to make sure your muscles are strong enough to do so. Dr. Metzl added
-
that the symphony is dead… but when the looked-for genius comes, it may be absolutely revived.”⏎A few years later, Elgar took it upon himself to be the “looked-for genius,” and the result was a staggering immediate success
-
We know that it is consistent to be able to live and not know. Some people say, "How can you
-
It is fascinating to find out how you came to be on this earth and who lived before you in
-
dimension to help others, give of one’s time, resources and energy, in aid of others.⏎- The Regulatory Dimension: The Regulatory drive indicates one’s drive to establish order, routine and structure. This motivation is to promote rules and policies,
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 have a very specific example of the last one I have, I have a very specific example of the last one I have... I have a very specific example of the last one I have... I have a very specific example of the last
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
have no underlying health issues, such as diabetes,” says Cohen. “And if it promotes weight loss, that can have great psychological benefits, which may spur you on to other healthy behaviors such as making regular exercise part of your life, too.”⏎
-
drinking alcohol.⏎Alcohol withdrawal occurs most often in adults. But, it may also occur in teenagers or children.⏎The more you drink regularly, the more likely you are to develop alcohol withdrawal symptoms when you
-
top causes of discoloration include food and drinks that have deep color pigmentation. in any food, the deeper or more intense the color, the more likely it will stain teeth. Teas, wines, soft drinks, sports drinks, berries
-
because fire ants prefer monocultures and will shy away from biodiverse areas.⏎By contrast, the more kinds of species that inhabit an area, the more likely it is that at least a few strains of plants or animals will be resistant to pests
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
- L5:2742 excites
- L4:2908 excites
- L5:1885 excites
- L5:2744 excites
- L4:635 weakly suppresses
- L4:2577 weakly suppresses
- L4:758 weakly excites
- L2:2769 weakly suppresses
- L4:398 weakly excites
- L5:102 weakly excites
- L1:949 weakly excites 'to' after 'agreed'
- L4:2308 weakly excites
- L5:110 weakly suppresses
- L4:1819 weakly excites
- L4:2470 weakly suppresses
- L2:2798 weakly suppresses
- L5:86 weakly suppresses
- L4:944 weakly suppresses
- L3:2614 weakly suppresses
- L4:1316 weakly suppresses
- L5:2217 weakly suppresses
- L5:1950 weakly excites unclear: 'Lorenz' and name fragments
- L2:1163 weakly suppresses
- L3:1457 weakly excites
Feeds later layers
- L8:369 strongly suppresses 'of' after 'learned' or 'because'
- L8:94 strongly excites
- L16:1153 strongly excites
- L13:909 strongly excites 'to' before a verb after a noun
- L7:681 excites 'from' after prevent or stop
- L15:1176 excites
- L9:226 excites 'to' starting an infinitive
- L19:385 suppresses
- L18:472 excites
- L7:138 excites 'to' after verbs like prove or time
- L8:723 excites
- L14:659 excites
- L18:715 excites
- L17:1592 excites
- L19:617 excites
- L17:472 excites
- L19:776 excites
- L20:480 excites
- L17:187 excites
- L17:1994 excites
- L7:8 suppresses
- L19:652 excites
- L19:2515 excites
- L20:241 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
- 把它
- /Images
- ��
- 将其
- ivid
- 把她
- 론
- stalk
- manufacture
- ifice
Suppresses
- weakest
- 不确定性
- gameState
- edido
- eso
- principio
- consistency
- etas
- ewitness
- ::~