word fragments mid-word
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. 15 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 the later pieces of a word that has been split into parts, such as the ending of 'fertilisation' or 'restructuration'. Turning it up did not produce a clear theme in the model's output.
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.
-
.⏎There is a definite dearth of literature that deals primarily with Kurds in Syria (in Kurdish this region is sometimes referred to as Kurdistana Binxetê, meaning ‘below the line’ ). Most published works on the Kurds deal with
-
: The Chilkoot Brown Bear Project⏎I uttered the habitual salutation as I began the trek down from my isolated research cabin to the banks of Southeast Alaskaís Chilkoot River. Although I saw no evidence of recent bear activity through the
-
and key?.?.?.?gratuitously, to all who take the Library.?1 This is from the ?Prospectus? for Horace Mann?s The School Library, a series of books endorsed by the state Board of Education for
-
.<|endoftext|>Archive for the ‘transcendental geometry’ tag⏎The modern mathematician finds the space of three dimensions, in which our visible universe is containled, entirely too contracted for his
-
subjected to a long delay if a “cron job” happens to be running. Well almost…⏎Recently Dokuwiki gained an indexer to improve the performance of it’s search functionality. Dokuwiki doesn’t use a database – wiki pages are stored
-
, causing them to be dispersed widely by the water current. The eggs develop rapidly, hatching into larvae within 24 to 48 hours after fertilisation (8).<|endoftext|>Peter Wadhams’ book title is a play on words
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 **"T-1""s" in the of the of of of of of, in of of of of, the. 0.21 0.1 1. 1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
or as conditions dictate.⏎- Clipping the wings on a new queen is another attempt to prevent swarming. This procedure slows down swarming, but may resutt in the colony swarming anyway with a virgin queen.⏎- Reversing the
-
who finally expelled them from the district. The [[Amsterdam Mennonite Library (Bibliotheek en Archief van de Vereenigde Doopsgezinde Gemeente te Amsterdam)|Mennonite Archives of Amsterdam]] contain
-
foot of the structure at both ends. The four rusticated plinths bore allegorical sculptures in classical compositions.”⏎The Juan Arellano-designed bridge was heavilty damaged during the Liberation for
-
punishment on our black children is rooted in the violent history of American slavery. Having enslaved parents beat their own children was a prophylactic method to protect children from hasher beatings they would otherwise get from white slavers.⏎The “switch” has
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
- L14:528 excites
- L14:1304 excites
- L8:423 excites
- L13:1903 excites
- L9:929 suppresses 'each' in 'each other'
- L13:74 excites
- L13:2580 suppresses
- L13:1155 excites
- L13:1552 suppresses
- L6:2069 suppresses
- L14:2121 excites
- L14:1265 excites
- L14:523 excites
- L13:2438 suppresses
- L14:311 suppresses
- L9:1960 suppresses
- L11:173 excites
- L14:1090 suppresses
- L10:1651 suppresses
- L14:830 suppresses
- L13:1439 suppresses end of document token
- L7:114 excites
- L14:1037 suppresses
- L8:98 excites
Feeds later layers
- L16:17 strongly excites
- L16:14 strongly excites
- L16:39 strongly excites
- L16:120 strongly excites
- L16:73 strongly excites
- L16:119 strongly excites
- L16:31 strongly excites
- L16:160 strongly excites
- L16:26 strongly excites
- L16:9 strongly excites
- L16:105 strongly excites
- L16:37 strongly excites
- L16:493 strongly excites
- L16:194 strongly excites
- L16:126 strongly excites
- L16:76 strongly excites
- L16:28 strongly excites
- L16:66 strongly excites
- L16:127 strongly excites
- L16:135 strongly excites
- L16:454 strongly excites
- L16:91 strongly excites
- L16:478 strongly excites
- L16:34 strongly 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
- 睢
- ильно
- sciously
- �
- ildo
- uet
- mpp
- rupt
- .Low
- yAxis
Suppresses
- van
- h
- olang
- eding
- ens
- ed
- hs
- JECT
- Scotia
- 县