"as well" after "as"
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. 6 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 words "well" in the phrase "as well as", where it adds something to a list. When turned up, the writing went off into numbers and math problems instead of the prompt's topic.
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.
-
mounts. The history of its inhabitants is inexorably linked with silver, this metal that provoked a demographic and economic boom in the city during three centuries, as well as its decline and poverty that came next when it became rarer. The richness
-
olytus was a thorn in Pontain’s side. His rigidity in practice and his arguments on the relationship between Jesus and God the Father, as well as his hatred of Pope Callixtus, lead to his schism with
-
. With their caravans composed of hundreds of camels, Semitic tribes ferried silk, cotton, precious stones and spices from China and India, as well as Arabian incense, all sold at exorbitant prices in Rome. Thanks
-
seen as definitive proof that widespread use of computers, phones and video games affects students’ capability to focus.⏎Even so, the researchers who performed the studies, as well as scholars who study technology’s impact on behavior and the brain, say the studies
-
apaskan hunting campsite and very little specific information was known about these camps. Another important consideration was that the site was buried and with this protective cover, as well as an apparent spatial dimension to the site, it appeared possible that activity specific areas
-
the lawlessness and violence of Japan around him, which was simultaneously going through the breakdown of the feudal system imposed by the Tokugawa shoguns, as well as the impact of the West. By 18<|endoftext|>One space or two
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 user is currently not logged in" (this is a command or action that should have been performed by the system). Please add or remove the action. This is a command to be performed by the system. Please add or remove the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a landscape digital photographer. It does not utilize the camera. The professional photographer takes pictures of natural scenes like beaches, hills, or falls. Modern modern technology has actually made it possible for expert landscape photographers to create pictures
-
paintings.⏎Sextus Empiricus' Outlines of Pyrrhonism is among the most crucial and influential texts within the historical past of Greek philosophy. within the calls for of cause Casey Perin
-
view of the discoveries, people and customs of the Pacific region.⏎A. Terre de Kerguelen, Appellee par M. Cook, Isle de la Desolation [on sheet with] Plan du Port Palliser, dans la
-
use of the phrases hope, love, dwelling, college, youngsters, group, help, and assist after the drills. The researchers thought of posting with these phrases within the aftermath of the drills to be indicative of getting excessive nervousness.⏎They included
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
- L5:164 suppresses
- L4:2073 excites
- L2:1 suppresses
- L2:363 suppresses
- L4:1918 suppresses
- L4:1617 excites
- L5:1909 suppresses
- L2:278 weakly excites
- L2:1127 weakly excites
- L5:1142 weakly suppresses
- L4:763 weakly excites
- L2:1472 weakly excites unclear: weak, scattered activations
- L4:50 weakly excites
- L2:1021 weakly excites
- L4:648 weakly excites
- L2:606 weakly excites
- L4:122 weakly suppresses
- L2:87 weakly suppresses
- L3:159 weakly suppresses
- L4:823 weakly suppresses
- L2:990 weakly suppresses
- L4:1797 weakly excites
- L2:35 weakly excites
- L1:1068 weakly excites
Feeds later layers
- L7:184 excites
- L9:110 excites
- L7:281 excites
- L7:1333 excites
- L7:114 excites
- L8:209 excites
- L7:3013 excites
- L8:1549 excites
- L9:1952 excites
- L7:840 excites
- L8:256 excites
- L7:130 excites
- L14:1074 excites
- L8:392 excites
- L8:694 excites
- L14:523 weakly excites
- L7:2887 weakly excites
- L8:1379 weakly excites
- L7:437 weakly excites
- L9:1454 weakly excites
- L16:1410 weakly excites
- L20:822 weakly excites
- L14:1614 weakly excites
- L8:2709 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
- ```
- 运动
- 符号
- 信息
- pec
- erm
- Find
- ox
- 運動
- 符
Suppresses
- ourke
- dać
- theless
- icens
- cern
- ớt
- liness
- etur
- isable
- ,Th