parenthesis introducing an acronym
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. 25 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 responds to the bracket that opens a short form right after its full name, such as the "(" in "Federal Emergency Management Agency (FEMA)". Turning it up made the writing fill up with the sound "pe" over and over, but the negative side also hits these brackets, so its role is not clean.
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.
-
, and gave glycerin-supplemented poultry feed to broilers that were 7 to 45 days old.⏎Dozier evaluated the apparent metabolizable energy (AME) intake—a standard measure of energy—for the study group and for a control
-
, for example, mentioned above as contributing to the dissolution of boundaries in education, can also be instrumental in reinforcing existing boundaries. The concept of Personal Learning Environments (PLEs), whether they extend control of existing learning platforms or enable the independence of
-
verse Childhood Experiences (ACEs). This team has developed a child ACE screening instrument in an effort to
-
with persistent breathing problems and other post-COVID complications.⏎The new method developed by KAUST — known as Deep-Lung Parenchyma-Enhancing (DLPE) — overlays
-
The Australian Government is making it easier for new migrants to learn and improve their English language skills by increasing the number of available hours through the Adult Migrant English Program (AMEP).⏎From today, migrants in Australia will
-
ussions, a multicenter, prospective diagnostic study was launched and led by the Advanced Research Core (ARC) of the Concussion Assessment, Research, and Education (CARE) Consortium. The consortium is a joint effort of the National Collegiate Athletics
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: what if the public input into the data and the data into the public input was dual? That is, the public input would carry the data and the data would carry the public input, such that both ends have a dual input and dual output
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
do Fogo stratovolcano - photo © 2013 Carole and Frederic Hardy⏎Fogo - Cha das Caldeiras - Pico Pequeno - photo © 2013 Carole and Frederic
-
potentially predict a disaster, usually in regards to weather, and how to measure the impact of a disaster immediately after its impact is felt. The Federal Emergency Management Agency (FEMA) often has questions like these. One indicator they informally use is
-
the National Cancer Institute and the CDC analyzed cancer incidence and follow-up information from nine cancer registries that are part of the Surveillance, Epidemiology, and End Results (SEER) program, which covers approximately 10% of the U.S
-
lets mostly smooth; plants 2-6 m tall var. arborescens (T.& G.) A. Gray. Nutlets slightly wrinkled or pebbly; plants 0.5-3 m tall⏎Moist
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
- L22:2948 strongly suppresses
- L22:109 strongly excites
- L24:2225 excites
- L23:1707 suppresses
- L22:285 suppresses
- L24:1585 suppresses
- L12:2194 excites
- L12:472 weakly excites
- L24:1623 weakly suppresses
- L9:337 weakly excites
- L4:2513 weakly excites
- L23:2829 weakly excites
- L24:235 weakly excites
- L21:44 weakly suppresses
- L13:249 weakly excites
- L23:54 weakly suppresses
- L6:2896 weakly excites
- L8:609 weakly excites
- L13:2600 weakly suppresses
- L8:1458 weakly excites
- L17:2930 weakly suppresses
- L22:2849 weakly suppresses
- L22:168 weakly excites
- L11:2138 weakly excites
Feeds later layers
- L27:178 strongly suppresses
- L27:273 excites
- L27:1188 excites
- L27:2083 excites
- L27:621 excites
- L26:2865 excites
- L27:796 excites
- L27:1059 weakly excites
- L27:1243 weakly excites
- L27:1525 weakly excites
- L26:869 weakly excites word before him or he
- L27:1563 weakly excites
- L27:1085 weakly excites
- L27:1334 weakly excites
- L26:1101 weakly excites
- L26:109 weakly suppresses
- L26:1356 weakly excites
- L27:2404 weakly excites
- L26:2102 weakly excites
- L27:1994 weakly excites
- L27:2268 weakly excites
- L27:909 weakly excites
- L27:2325 weakly suppresses
- L26:504 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
- (pe
- PE
- pe
- _pe
- Pe
- Pe
- pe
- PE
- -pe
- CE
Suppresses
- 公司章程
- Pandora
- 盱
- ImageSharp
- forth
- 工商银行
- AxisAlignment
- 磺
- istrator
- 打球