unclear: weak scattered tokens
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. 9 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 gives only weak signals on a scattered mix of words, such as times like "4:30 p.m.", "no" and "not". Its negative signals are also unclear. Turning it up or down did not produce a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
half, the winds picked up dramatically with gusts reaching nearly 44 mph as the storm rolled to the southwest.⏎By 4:30 p.m., the storm’s outflow boundary—where the the thunderstorm’s cooled air is
-
' nearly permanently.⏎Sometimes we allow the connections to be 'forced' on us. For example, if you were to be asked where you were last Wednesday at 15.00 hours, chances are you wouldn't immediately recall. But
-
ans and journalists gathered to inaugurate the city’s new air terminal on May 3, 1961, the significance of the event was lost on no one. Though Atlanta boosters already considered their home the “capital of the South
-
relationship existed between the paired data. Furthermore, regression analysis was used to identify the exact nature<|endoftext|>Coloring pages - Edupics.com⏎All videos (0)⏎Amusement park (15)⏎Children's rights (14)⏎Space exploration
-
what is left when something happens but does not completely undo itself. Meaning that is an experience, but it cannot fully get erased
-
"I get up at five in the morning and start the day by milking the goat and taking the animals out to pasture. Only after that do I go to school," said Gyorgy Radavoi, a 1
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 thinking process of the user of the previous steps, the thinking process of the user of the previous steps, the thinking process of the user of the previous steps, the thinking process of the user of the previous steps, and so on.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Alberta’s international merchandise exports last year. If we exclude oil and gas extraction and petroleum refinery exports, its share rises to 36%.⏎- Alberta exports the third largest amount of agricultural
-
is appreciated. His temperature is 99.6°F. He has no leukocytosis and CRP and ESR are normal. Radiographs of the right leg are seen in Figure A. What is the
-
its natural fleece (not shorn).⏎Chest measurement shall be done with a soft but strong measuring tape just behind the shoulders. The person measuring shall always stand on the left side of the sheep, slide the end of the tape down the right
-
ini protested that he needed to see for himself. At the seance, Houdini saw through the deception and called “Margery” on it, much to the annoyance of Bird, who angrily resisted exposing her con game. The arguments that
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
- L8:191 suppresses
- L8:112 excites
- L7:512 excites
- L6:240 excites
- L7:2158 excites
- L6:289 suppresses
- L7:736 suppresses
- L6:1433 suppresses
- L8:1205 excites
- L7:586 suppresses
- L7:186 excites
- L7:298 suppresses
- L5:1128 excites
- L6:83 suppresses
- L7:347 suppresses unclear: mid-sentence function words
- L8:294 suppresses
- L7:2750 suppresses
- L4:297 weakly suppresses
- L7:177 weakly suppresses
- L6:936 weakly excites
- L5:585 weakly excites
- L5:2008 weakly excites
- L5:2163 weakly suppresses
- L7:84 weakly suppresses
Feeds later layers
- L10:766 excites
- L10:1073 excites
- L10:1863 suppresses
- L14:704 excites
- L16:1193 excites
- L14:995 excites
- L15:1333 excites
- L11:459 excites
- L10:1830 excites
- L13:86 suppresses
- L12:1467 excites numbers after dates and figures
- L14:924 excites
- L13:423 excites
- L14:1323 excites
- L19:1065 suppresses
- L13:1203 excites
- L15:2091 excites
- L13:858 excites
- L12:1426 excites
- L19:2653 excites "most" meaning "very" or "the most"
- L13:1343 excites
- L10:541 suppresses
- L13:1725 excites
- L13:2322 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
- 已久
- 十条
- ://${
- 很高
- 非常高
- 极高
- 很高的
- nowhere
- -date
- HasBeenSet
Suppresses
- sons
- jango
- 这两年
- Registr
- eft
- зал
- Tail
- itches
- Registr
- TAIL