first word of a proper name
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. 13 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 first word of a name that is about to be completed, such as 'Civil' in 'Civil War', 'Occupational' in 'Occupational Safety' or 'Storm' in 'Storm Water Credit'. Turning it up did not produce a clear theme, so this is only a guess from the examples.
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.
-
used in the production of fuel cells. A very simplistic description of the technology is that a fuel, typically hydrogen, is oxidized in the first segment of the fuel cell to create positively charged ions and
-
and vocabulary. All the spellings on this worksheet end in 'ag'. Students learn the spellings and then answer the questions. This worksheet was created by www.primaryleap.co.uk⏎(there is an issue with the thumbnail, showing the
-
plan, the city's Sustainability Advisory Board identified a need to focus on education and outreach on shoreline restoration, and to encourage private property owners to use incentives like the Storm Water Credit Program, said Elizabeth Williams, associate planner for the city.⏎RELATED:
-
materials may expose workers to health risks. Developed countries regulate manufacturing activity with labor laws and environmental laws. In the U.S, manufacturers are subject to regulations by the Occupational Safety and Health Administration and the United States Environmental Protection Agency. In Europe, pollution
-
this piece of legislation threatened the very existence of the Society as it was feared that the entire organisation, including its funds and lands, would be taken over by the Poor Law Commissioners. However, this never occurred as the Society was a voluntary organisation which
-
Philosophy recognized this void and secured the support of the College of Humanities and Social Science to begin developing the Center. After nearly a year of planning and designing, the Civil War Center became a functioning unit within the college. Thus, the mission and purpose
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: need to do this also add to the list of this also add... so do this also add... this also add... so do this also add... so do this also add... this also add... so do this also add... so do
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
battles.⏎An effort to improve the Luftwaffe support<|endoftext|>Extra 10% off sitewide + Free shipping over $150 with code: STOCKUP Details⏎Happy Homeschool Month! The whole month of January we will have
-
2008, mostly through birth or breastfeeding from an HIV-infected mother. Many regions of the world are gaining increased access to complex antiretroviral drug regimens for preventing HIV transmission from a mother to her child. However,
-
⏎4 | 59⏎5 | 01⏎6 | 3⏎9 | 4⏎The way the app generates the stem-and-leaf plot for your dataset may be different.⏎The stem-and-leaf plot allows us
-
.4 cigarettes a day – the lowest daily cigarette consumption since the series began, the peak being in 1976 at
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
- L6:157 suppresses "of" after process or action nouns
- L8:928 excites
- L12:1345 suppresses
- L10:1317 excites
- L12:2753 excites
- L12:2056 excites
- L5:2594 suppresses
- L8:2768 excites
- L12:2663 excites
- L11:2163 suppresses
- L9:171 suppresses
- L9:194 suppresses
- L8:1213 excites
- L10:235 suppresses
- L9:277 suppresses
- L10:579 suppresses
- L12:940 suppresses
- L12:573 excites
- L12:996 suppresses
- L12:1853 excites
- L12:973 suppresses 'the' in 'hit in the body part' idioms
- L12:1974 excites
- L11:2467 suppresses
- L7:2257 weakly suppresses
Feeds later layers
- L14:304 excites
- L16:1508 excites
- L16:189 excites
- L14:1422 suppresses
- L16:322 excites
- L15:168 excites
- L15:2239 suppresses
- L17:641 excites
- L15:227 excites
- L14:23 excites
- L16:458 excites
- L15:2 weakly excites
- L19:76 weakly excites
- L15:6 weakly excites
- L14:211 weakly excites
- L16:41 weakly excites
- L15:613 weakly excites
- L15:9 weakly excites
- L14:752 weakly excites
- L14:2312 weakly suppresses
- L15:42 weakly excites
- L15:466 weakly excites
- L17:562 weakly excites
- L14:25 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
- 一览
- _bal
- rophic
- доволь
- 所致
- iffs
- .only
- quals
- сет
- thankful
Suppresses
- yah
- lend
- lander
- estre
- LOTS
- endor
- terra
- atter
- oble
- TEX