closing quote marks at the end of 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. 11 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 at the end of a line of dialogue, where a sentence ends with a closing quotation mark and a line break. Turning it up did not make the model's output clearly about speech or quotes.
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.
-
drawing a picture of.⏎"God," he said.⏎"Why Johnny," she said, "you can't do thatbecause nobody knows what God looks like."⏎"They will when I'm finished," he assured her.⏎That story presents concret
-
-clad horsemen.⏎“I’ll wager that’s General Lee,” said the captain, with a glance at me. “Let us hope things turn out all right.”⏎As the two Federals neared the spot, a man arose and they exchanged
-
You quite shock me; if you mean a fling at the slave-trade, I assure you Mr. Suckling was always rather a friend to the abolition.”⏎“I did not mean, I was not thinking of the slave-trade,” replied Jane
-
already mention that you might have a rare disease?”⏎That was the first shock. “I might have what?”⏎She said: “A rare disease that is called Alpha 1 Antitry
-
Do you see the cell phone as sort of an arriviste, as someone that sort of came along and stole the thunder of the payphone?” Rocca asked.⏎“It stole its relevance. And it made communication so simple and so trivial even
-
“I sometimes think the western countries that are financially supporting this degradation process have no idea what they are destroying. There is nothing in Europe remotely like this river system.”⏎From the mountains of Greece where it is known as the Aoös, the V
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 been thinking about lately, and I have been thinking about some of the most important things for me, like me, like, I have been thinking of me, like, and I have been thinking of me, like, and I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
his eyes up to the general. His mind and body seemed two separate things and the young soldier couldn’t stand at attention or even offer a salute. “I-I guess I can’t take it, sir,” Kuhl said.⏎The general’s
-
, and I had to have it removed. I really love birch trees, but I wonder if it's worth it to try planting one again.⏎A Yes, try again. Chances are good that a new birch will live long enough
-
the most common types of tomatoes available. They can be meaty like a beefsteak, but always juicy.⏎Vesuvio Piennolo<|endoftext|>Thank you to DadsWorksheets.com for providing a VARIETY of math practice sheets
-
might have what?”⏎She said: “A rare disease that is called Alpha 1 Antitrypsin Deficiency.”⏎I was confused. “What? What is this? How can I cure it? What do I do?” I
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
- L2:3037 suppresses
- L6:2259 weakly suppresses
- L8:16 weakly suppresses
- L10:2351 weakly suppresses science news body text
- L8:10 weakly excites
- L10:2771 weakly suppresses
- L10:1878 weakly excites
- L9:933 weakly suppresses
- L10:624 weakly excites
- L8:2057 weakly suppresses
- L4:1687 weakly excites
- L0:1670 weakly suppresses
- L2:2041 weakly suppresses
- L5:364 weakly excites
- L4:310 weakly excites
- L10:2278 weakly suppresses
- L1:2444 weakly suppresses
- L4:2994 weakly suppresses
- L0:2192 weakly suppresses
- L6:1085 weakly excites
- L2:2502 weakly excites
- L10:36 weakly excites
- L10:2227 weakly suppresses
- L8:471 weakly suppresses
Feeds later layers
- L15:3036 excites
- L15:2027 weakly excites
- L15:2012 weakly excites
- L17:1266 weakly suppresses
- L12:67 weakly excites
- L13:2559 weakly suppresses
- L21:2349 weakly suppresses
- L23:2166 weakly suppresses
- L25:1044 weakly excites
- L13:2792 weakly suppresses
- L24:793 weakly suppresses
- L24:1932 weakly suppresses
- L12:2444 weakly suppresses
- L12:2413 weakly suppresses
- L13:2020 weakly excites
- L13:845 weakly excites
- L18:2123 weakly suppresses
- L19:152 weakly suppresses
- L18:1657 weakly suppresses
- L19:2013 weakly excites
- L24:1758 weakly suppresses the 'Ar' syllable inside names
- L16:3043 weakly suppresses
- L24:1046 weakly suppresses
- L20:2748 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
- IFIER
- amide
- uating
- ائم
- ipients
- phans
- 玑
- uity
- lsi
- ossa
Suppresses
- TEM
- nore
- createElement
- ıklı
- VE
- raq
- ᾶ
- :"; ⏎
- 你也
- 榄