the word 'verb' in grammar explanations
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. 12 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
The neuron fires most on the word 'verb' when a text explains where a word comes from or what part of speech it is. Turning it up made the model's replies wander into repetitive text, so the steering gives no clear sign of the same 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.
-
Origin late Middle English: from Latin attentio(n-), from the verb attendere (see attend). Nonetheless, one shouldn’t confuse a job application kind
-
Egyptians will know 3 that I am the Lord.” So this is what they did. 4⏎|NET © Notes||⏎1 tn In this place the verb חָזַק (hazaq) is used; it
-
(26 km²), 0.92%⏎44/sq mi (17/km²)⏎|Time zone||Central: UTC-6/-5|⏎Cherokee County is a county located in the U.S.
-
surname harks back to the jaggers of yesteryear - hardy men who carried heavy packs - they were pedlars and hawkers. The verb "to jag" meant to pack or to carry. When good roads were built
-
att, David M., "Music Listening as Therapy" (2015). Loma Linda University Electronic Theses, Dissertations & Projects. Paper 294.⏎Loma Linda University Electronic Theses and Dissertations⏎
-
not far from Kobe; he died in the year 737.⏎In the fourth line nagashi may be taken as the adjective 'long', or the verb 'to drift along'; and yo may mean either 'night' or 'life
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 class of students, and I have a class of students, and I have a class, and I have a title, and I have a new one, and I have a new one, I have a new one, I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
50 percent, a candid admission of the challenges the Tennessee schools faced.⏎West Virginia, New York, Nebraska, and Delaware also strengthened proficiency standards, while New Mexico, Washington, Hawaii, Montana, and Georgia lowered the bar.⏎Uneven at
-
this article we will mainly discuss the significance of fasting insulin with respect to insulin dependent Diabetes Mellitus or diabetes type 1 and also non insulin dependent diabetes or diabetes type 2.⏎Diabetes is worldwide in distribution and the incidence of both types of
-
health insurance. About 44% of these people were white, 33% were Hispanic, 15% were black, 5% were Asian/Native Hawaiian or Pacific Islander, and 3% were of another race
-
of color? In eleven states, about a third or more of black women were poor (IA, IN, LA, ME, MI, MN, MS, SD, WI, WV, and, VT). Roughly a third of Hispanic women
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
- L11:232 strongly suppresses
- L11:183 strongly excites
- L11:321 strongly excites
- L11:435 strongly suppresses
- L11:284 suppresses
- L11:262 excites
- L11:235 suppresses
- L11:2429 excites
- L10:537 suppresses
- L10:542 suppresses word completing a technical term
- L10:178 excites
- L11:800 excites
- L10:316 excites
- L10:414 suppresses
- L10:711 suppresses
- L10:211 suppresses
- L10:399 excites
- L9:305 suppresses
- L4:190 excites
- L10:361 excites
- L11:740 suppresses
- L10:209 weakly suppresses
- L4:245 weakly suppresses
- L9:168 weakly excites
Feeds later layers
- L13:219 excites
- L13:177 suppresses
- L17:183 suppresses
- L15:112 suppresses
- L16:501 excites
- L13:632 excites
- L13:258 suppresses
- L13:224 excites
- L15:205 weakly excites
- L16:843 weakly suppresses
- L16:596 weakly excites
- L16:427 weakly suppresses
- L13:375 weakly excites
- L13:953 weakly excites
- L13:2504 weakly suppresses
- L15:471 weakly excites
- L13:343 weakly excites
- L25:1260 weakly excites
- L14:1451 weakly suppresses
- L17:536 weakly excites
- L15:183 weakly excites
- L17:64 weakly suppresses
- L16:563 weakly excites
- L16:687 weakly 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
- ioni
- ivr
- contro
- imes
- ione
- iv
- }`}⏎
- íveis
- eful
- Dez
Suppresses
- партнер
- コード
- парт
- 臊
- IFY
- év
- 篇
- cname
- товар
- 第一