words after 'what' and 'go through' in speech
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 mixed: they peak on different tokens. 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 small words that come right after a verb or 'what', such as 'through' in 'go through' or the 's and 're in 'what it's' and 'what you're'. It is most active in casual, spoken-style sentences. Its negative side, on words like 'says' after a person or group, is a different pattern.
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.
-
the events of that tragic year of 1941, but our memory should not allow us to forget what the Soviet people had to go through in those blood-filled years of 1941-1945
-
."⏎- “Godfather doesn't like being told what to do by the enemy.”⏎- “Godfather needs an airfield.”⏎Mark wanted to know what it’s called when someone talks like this. And he wondered if he was alone in thinking
-
of the ovaries in a woman is a natural process that every female has to go through at some point in her life. By medical standards, this time is fixed during
-
whatsoever. It's completely dark. That is what we're looking for and that is what makes this area so incredibly special."⏎The Namib
-
’ve heard this word many times over the last few years, but unless you have actually participated in a detoxification program, you probably don’t really understand what it’s all about. Given the somewhat controversial nature of this topic, as well as the
-
lived in. That was the social contract before the New Deal. There was no Medicaid for the poor, no subsidized housing or home mortgage interest deduction, no unemployment insurance, no Social Security for the disabled or the elderly. Life was nasty, brut
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: the problem of a couple. So, for the first time, I have been thinking about the problem. What is the answer to the question: "What is the sum of the two numbers 3 and 2? Let me know the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and positive psychology is impacting the behavior of more than 70 boys with serious behavioral or criminal issues.⏎Salekin and his team have developed a system that he says has been more effective at reaching troubled youth at the facility than typical methods, like
-
tide laps across them, back and forth, every day.”⏎One set of markings came from a two-legged, meat-eating dinosaur that researchers say was about the size of a car. They say its footprints have three toes
-
a book or looking it up on a webpage. In particular – I will be focusing on the Romans and the Greeks as these are especially interesting, and in my opinion, more factual. We can learn a lot from visiting sites, but we can
-
to a debate that we are likely to have in the coming decade, a debate about how we know the past and how we look at the written record that I suspect will be of interest to literary scholars and historians alike. It is a philosophical debate
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
- L5:2622 weakly excites
- L5:1030 weakly excites
- L4:2187 weakly excites
- L2:627 weakly suppresses
- L1:2605 weakly excites
- L5:1463 weakly excites
- L4:179 weakly suppresses
- L4:1870 weakly excites
- L4:929 weakly excites
- L5:657 weakly excites
- L0:1371 weakly excites
- L4:2564 weakly suppresses
- L3:265 weakly suppresses
- L0:1847 weakly excites
- L4:153 weakly excites
- L4:19 weakly suppresses
- L2:2402 weakly excites
- L5:198 weakly excites unclear: weak "and" in lists
- L4:118 weakly excites
- L5:1495 weakly excites
- L3:3 weakly suppresses
- L1:2328 weakly suppresses
- L1:1590 weakly suppresses
- L5:619 weakly suppresses
Feeds later layers
- L17:1192 suppresses
- L8:2127 suppresses
- L7:2601 suppresses
- L15:566 suppresses
- L8:1847 suppresses
- L18:2153 weakly suppresses
- L12:2966 weakly excites
- L8:507 weakly suppresses
- L8:502 weakly suppresses
- L14:752 weakly suppresses
- L17:1756 weakly suppresses
- L15:452 weakly suppresses
- L11:2106 weakly suppresses
- L8:2822 weakly suppresses
- L14:647 weakly suppresses
- L7:357 weakly suppresses
- L14:1124 weakly suppresses
- L23:1196 weakly excites
- L21:2408 weakly suppresses
- L19:640 weakly excites
- L21:496 weakly suppresses
- L14:500 weakly suppresses
- L12:2063 weakly suppresses
- L16:2315 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
- 庥
- <%=
- ��
- <?=
- %;">⏎
- %;⏎
- -<?
- <?=
- "><?
- Altern
Suppresses
- obuf
- gebn
- elocity
- iske
- 体育彩票
- olut
- ynomials
- وسط
- pository
- urname