'to' after 'not' (not to do)
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. 4 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 fires on the word 'to' when it comes right after 'not', as in 'not to disturb' or 'not to go out'. Turning it up did not produce clearly related writing; the outputs wandered off in different directions.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
lightly so as not to disturb the soil. Continue adding water until the excess flows from the bottom drainage holes
-
which helps you lose weight. If calories consumed in food are less than calories spent during exercise and other daily activities, you will lose weight. It is impossible not to! Many people have learned how to lose belly fat by starting a running program.⏎
-
in the country -- in that respect, it’s a pretty average state.⏎Largest Cities in West Virginia⏎West Virginia is one of the few US states not to have a city with more than 100,000 residents.
-
so that she is certain not to go out of the lines while another little girl with the same princess coloring pages scri
-
regulations for businesses. Its leader, deputy leader and finance spokesperson, Tim Farron, enjoys a strong base of support in the<|endoftext|>Many of us are inclined not to talk about things that upset us. We try to put a lid on our feelings
-
past”⏎“Look forward, not back”⏎All said by people who seemingly think that experiences of trauma can be consciously resolved, if we just try hard enough not to think about them, as if it were a choice to be impacted by our worst
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 4-day event. The first day, I do a task that involves a game, and the second day also, I have a different, more complex task. The goal is to get a full salary. However, the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
together⏎- Singing songs⏎- Nursery Rhymes⏎- Question and answer board games⏎- Te Reo Singalong with Sharon Holt⏎- Listent to stories – a collection of free children’s audio books. All written and produced in
-
make less than the top 25 percent of high school graduates.⏎This alternative reading of the figure helps to explain why many young people have not responded so enthusiastically to the “market signals” about college. On average, college pays off, but
-
to produce a treatment response as olanzapine. Both positive symptoms (psychosis) and negative symptoms (blunted emotional response, reduced motivation) responded better to clozapine. John H. Krystal, M.D., Editor of Biological
-
. Curiously, it was the king of Egypt, and not God, or the Israelites themselves, who coined the phrase, "Israelite People." Acting out of panic, Pharaoh wanted to do away with the Hebrews, when he
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
- L3:1866 weakly suppresses
- L3:122 weakly excites
- L3:59 weakly excites
- L3:421 weakly suppresses
- L3:370 weakly suppresses
- L1:1507 weakly excites
- L3:456 weakly suppresses
- L0:1372 weakly excites
- L1:2438 weakly excites
- L0:2721 weakly excites
- L3:115 weakly suppresses
- L3:272 weakly excites
- L2:1275 weakly excites
- L3:91 weakly excites
- L0:966 weakly suppresses
- L1:1973 weakly suppresses
- L1:318 weakly excites
- L0:2726 weakly excites
- L2:1708 weakly suppresses
- L3:2212 weakly excites
- L0:2280 weakly suppresses
- L3:71 weakly excites
- L1:1051 weakly suppresses
- L0:1572 weakly excites
Feeds later layers
- L18:2419 weakly excites
- L15:1622 weakly suppresses
- L23:217 weakly suppresses
- L18:141 weakly suppresses
- L8:723 weakly excites
- L18:2755 weakly suppresses
- L5:1958 weakly excites
- L26:836 weakly suppresses
- L15:2198 weakly excites
- L17:187 weakly excites
- L17:2097 weakly excites
- L15:397 weakly excites
- L13:1892 weakly suppresses
- L9:704 weakly suppresses the zero completing 10 or 100
- L6:439 weakly excites
- L24:1093 weakly excites
- L7:681 weakly excites 'from' after prevent or stop
- L14:2455 weakly excites
- L17:2466 weakly excites
- L24:2863 weakly excites
- L5:53 weakly excites
- L20:1652 weakly excites
- L20:530 weakly suppresses
- L6:3070 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
- .tf
- "#
- ук
- 到时候
- shade
- 选用
- 去
- nữa
- .Xr
- tensorflow
Suppresses
- otive
- loyment
- entire
- oting
- ritch
- alling
- aly
- antage
- oly
- abble