table pipes and 'non' before a hyphen
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. 18 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 mostly lights up on the vertical bar character used to draw table rows in plain text. Its strongest negative cases are the word 'non' before a hyphen, as in 'non-climatic'. The steering results are scattered and do not show a clear theme.
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.
-
object identifiers) is encouraged.⏎Citing the date when an online work was consulted is now optional.⏎Selected examples of MLA 7 vs. MLA 8⏎|MLA 7||MLA 8|⏎Article from a library⏎|
-
look at some examples of deferred revenue and how to determine if something is deferred revenue<|endoftext|>Your location is outside the National Library of Australia. Resources you can access:⏎|Show by title|⏎ACT Heritage Library: The A.C.T. electoral rolls for
-
), and freshwater fish (2,000 species) than any other continent.⏎What continent has the largest animals?⏎List of largest land carnivorans⏎|Rank||Common name||Native range by continent|⏎|1||Polar bear
-
PPL) and cross polarised (XPL) . As an example, in a Cornubian granite, the following minerals have the following basic properties:⏎|Quartz||Dull!
-
and Schedule⏎|4 years through 8 years||One or two dosesa, 0.
-
ease of use and because a simple application tends to grow in complexity with time.⏎The global pattern to follow is MVC and the layering:⏎(View & Controller)⏎|View||HTML templates usually generated using a template engine|⏎|Input Controller||Filter
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- 0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to adapt over the long term. People have considerable knowledge and skills to adapt, often using innovative new solutions, but their ability to respond depends on a host of non-climatic factors that increases people’s vulnerability, such as conflicts that restrict mobility
-
as the violence and scope of protests decline. As the protests begin to lessen or become less intense, non-protesters will feel safer leaving their homes.⏎"There are other possible explanations for
-
job with just one subunit.⏎The authors stress that they are not arguing against natural selection as a process; they say rather that it can be aided by “non-ad<|endoftext|>Poet and writer Langston Hughes, famous for his elucidations of
-
akers, whether in our personal or in our political lives. Yet, few have the skills or ideas to create the conditions in which peace can result. Courses in non-violence are rarely taught in schools, and non-violence is rarely used by
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
- L6:1751 excites
- L9:3065 excites
- L2:668 weakly suppresses
- L16:38 weakly suppresses
- L17:157 weakly suppresses
- L1:516 weakly excites
- L0:626 weakly excites
- L1:339 weakly suppresses
- L16:282 weakly excites
- L15:129 weakly suppresses
- L16:79 weakly suppresses
- L16:1869 weakly excites
- L15:122 weakly suppresses
- L15:115 weakly excites
- L1:2629 weakly suppresses
- L4:2819 weakly suppresses
- L17:998 weakly excites
- L6:1063 weakly excites
- L15:89 weakly excites
- L13:1229 weakly excites
- L4:681 weakly suppresses
- L16:118 weakly suppresses
- L4:1326 weakly excites
- L7:78 weakly excites
Feeds later layers
- L19:2429 strongly suppresses
- L23:864 suppresses
- L19:1053 suppresses
- L23:1175 suppresses
- L20:1064 suppresses
- L19:792 excites
- L27:752 suppresses
- L22:494 suppresses
- L26:459 suppresses opening bracket before an acronym
- L25:149 suppresses
- L21:378 suppresses
- L26:526 suppresses
- L27:2159 weakly suppresses
- L21:118 weakly suppresses
- L19:1305 weakly excites unclear: verbs after a noun or 'that'
- L26:596 weakly suppresses
- L23:676 weakly suppresses the 'm' in a.m. and p.m
- L27:1459 weakly suppresses
- L21:2279 weakly excites
- L24:266 weakly suppresses
- L25:1030 weakly suppresses
- L27:807 weakly suppresses
- L27:83 weakly suppresses
- L26:771 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
- iment
- ACHE
- ọng
- ATIONAL
- fingert
- NCY
- าย
- edly
- finder
- ỡ
Suppresses
- -
- pat
- -C
- disc
- ť
- -M
- metric
- -c
- −
- atics