'and' joining two full clauses
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. 19 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 lights up on the word 'and' when it links two complete statements, such as 'this browser will no longer be supported and you may experience issues'. When it was turned up, the model fell into repeated filler words like 'is is is', and when turned down it rambled on with long, wandering sentences.
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.
-
not support the requirements of modern web applications and pose significant security risks. After May 1, 2017, this browser will no longer be supported and you may experience issues or not be able to access the site at all. We
-
the recipient at another branch of the same bank. These items are not cleared between institutions and therefore are not represented in statistics of payments exchanged between financial institutions in the clearings
-
, other sources of individual variation, or environmental conditions.⏎You have requested a machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Neither BioOne nor the owners
-
a distance of about 20 km from Una town and is one of the most visited pilgrimage sites in the region. The temple is dedicated to Lord Shiva and is surrounded by beautiful natural surroundings, making it an ideal place for meditation and spiritual
-
fourth-most populous county. Its county seat and largest city is Ogden, the home of Weber State University. The county was formed in 1850 and named after the Weber River, which in turn was named for John Henry Weber (
-
fly. Despite your best efforts, sometimes the way you care for the gardenia may encourage whitefly infestations.⏎Whiteflies resemble tiny grains of white rice and tend to congregate on the underside of the gardenia’s foliage. They thrive
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: "If I, with the help of a new tool, that which I previously unable to do, and which I previously unable to do with a new tool, but with a new one, which I now able to do, which I now
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a common food source for many Native American tribes including but not limited to the Iroquois, Algonquin, Ojibwa
-
Athenae (see Athens).⏎(a.) Of or pertaining to Athens, the metropolis of Greece.⏎- (letter.) A native
-
place in a relatively small area of the Levant (which, in itself, is only a tiny part of what we today call the Middle East). Between (and including) the period of the Judges and the emergence of Israel as a kingdom,
-
6, the Coalition cabinet was authorized by the crown to present the Szell-Korber tariff to the Chamber in the form of a Hungarian autonomous tariff distinct from but identical with the Austrian tariff.⏎By an amendment of 1877,
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
- L16:376 suppresses 'or' joining alternatives
- L13:131 excites 'and' joining parallel phrases
- L18:2345 suppresses
- L18:1047 suppresses
- L10:74 excites
- L17:447 excites
- L15:727 excites
- L9:92 suppresses "and" or "or" joining two items
- L12:2141 weakly excites
- L18:207 weakly suppresses
- L14:1381 weakly excites
- L14:2820 weakly suppresses "and" or "or" linking two counted items
- L15:24 weakly excites
- L10:1706 weakly suppresses
- L12:869 weakly suppresses
- L14:1209 weakly excites
- L17:35 weakly suppresses
- L4:57 weakly suppresses
- L18:509 weakly suppresses
- L7:122 weakly excites
- L4:1999 weakly suppresses
- L18:445 weakly suppresses
- L17:1044 weakly suppresses
- L18:693 weakly excites
Feeds later layers
- L22:228 strongly excites
- L25:9 strongly suppresses
- L23:30 strongly suppresses
- L23:127 strongly suppresses
- L20:586 strongly excites
- L20:349 strongly suppresses
- L27:604 strongly excites
- L26:318 strongly excites
- L23:1013 strongly excites
- L21:262 strongly suppresses
- L26:53 strongly suppresses
- L23:320 excites
- L21:930 excites
- L20:162 excites
- L20:974 excites
- L21:1521 excites
- L27:804 excites
- L22:518 excites
- L27:565 excites
- L23:220 excites
- L22:1425 suppresses
- L27:321 excites
- L22:479 excites
- L27:1291 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
- would
- could
- can
- should
- will
- may
- must
- might
- are
- is
Suppresses
- olate
- ending
- leaving
- ORS
- abling
- UGIN
- nable
- ง
- meaning
- Ỏ