unclear: list commas and mixed 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. 18 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 a mix of things, including commas in lists and a few random words, and no single theme stands out. It is hard to describe in plain terms. Turning it up made the writing repeat words like "the system" or "the statement".
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.
-
the index. The index starts from 0 and ends at length of the array - 1.⏎i.e. the index of the first element is 0, index of the second element is 1… etc.⏎To print the
-
plain that runs from Mexico and southern Texas through Florida and along the Atlantic as far north Maryland and Delaware. It is a year-round bird within that territory but migratory farther north and farther inland. Its occurrence is also spotty farther inland because of
-
work is the "Historia General de los Hechos de los Castellanos en las Islas y Tierra Firme del Mar Océano" (General History of the deeds of the Castilians on the Islands and Mainland of the
-
to 34-year-olds had a college degree, compared with 35 percent of 35- to 44-year-olds and 38 percent of 45- to 54-year-olds.⏎Veterans from Gulf
-
Tendon Tears. These tendons are:⏎Quadriceps Tendon: The quadriceps muscle is a combination of four muscles which are the vastus lateralis, vastus medialis, vastus intermedius, and the rectus femoris
-
pronouns in Spanish agree in number and gender with the nouns they replace. In the following example, hijos is a masculine, plural noun so the masculine plural direct object pronoun los is used to replace it in the second sentence.Son muy buen
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 one that the one that the one that the one that the one that the one that the one that the one that the one that the one that the one that the one that the one that the one that the one that the one that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Dan Ferguson, Gregg Garfin, Mónica Ramirez-Andreotta and Connie Woodhouse designed the Transdisciplinary Environmental Science for Society (TESS) Program. The following parts have been prepared:⏎Part 1: Fundamentals of Transdisciplinary Research (
-
in many aspects of our lives including food quality and safety. However, mounting evidence shows that accumulation of plastic waste in oceans has detrimental effects on marine wildlife. The following infographics help visualize trends in waste generation, including plastics, for 19
-
influence on the attitudes of our time.⏎In a recent interview, DiFonzo talked about his research and his fascination with this particular form of communication. The following is an edited version of that interview:⏎City: What got you interested in rumors
-
in AutoCAD and allow engineers, clients, and the public to better understand a certain design. AutoCAD is an essential visual communication tool for civil engineers.⏎The following instructions
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
- L17:1160 suppresses unclear: inside numbers and names
- L16:2169 suppresses
- L17:2011 suppresses
- L16:1917 suppresses
- L17:1221 excites
- L14:1674 suppresses
- L15:1131 suppresses
- L14:1268 excites
- L16:2638 excites
- L8:78 excites
- L13:2560 suppresses
- L10:1062 excites
- L15:860 excites
- L4:270 excites
- L14:477 excites
- L15:2274 suppresses
- L15:470 excites the word 'this' after a preposition
- L17:1453 excites
- L3:16 excites
- L14:1254 suppresses
- L11:392 suppresses
- L14:1078 weakly suppresses
- L14:99 weakly excites
- L14:1141 weakly excites
Feeds later layers
- L19:1462 strongly excites
- L19:1737 excites
- L19:2384 excites
- L19:222 suppresses
- L19:2199 excites
- L19:2072 excites
- L19:1137 excites
- L19:1022 excites
- L19:687 excites
- L20:1988 weakly excites "the" before a topic noun
- L19:2586 weakly excites
- L20:1983 weakly excites
- L19:653 weakly excites
- L20:867 weakly suppresses
- L21:1119 weakly excites
- L19:1444 weakly suppresses
- L19:1557 weakly excites
- L19:1543 weakly excites
- L20:1229 weakly excites
- L24:2303 weakly suppresses capital G starting names
- L19:988 weakly excites
- L19:2673 weakly excites
- L20:2741 weakly suppresses
- L20:1458 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
- скор
- iasco
- apsulation
- ")] ⏎
- かり
- ) ⏎
- quisition
- 骈
- arcy
- sto
Suppresses
- 罄
- 度
- ак
- ListComponent
- ness
- (hdc
- elan
- ivant
- aced
- lined