'of' after a category noun
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 fires on the word 'of' when it follows a word for a kind or count of things, like 'types of', 'species of' or 'classes of'. Its weakest side shows up in phrases like 'all of', and turning it up did not produce any clear change we could tie to this habit.
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.
-
high altitudes are less than ideal for reptiles. These animals are ectothermic, meaning they gather body heat from their surroundings. There are only three species of reptiles in the park, with a fourth suspected to be present.⏎The common
-
.⏎Since 2019 hundreds of trees and shrubs have been planted by the group close to the Bucketts. Koalas eat many different species of eucalypts
-
.5km long, which is the same as 1,517 tyrannosaurus rexes laid end to end⏎- 10 species of bat are regularly seen in Bridgend County Borough<|endoftext|>Measurement, as a method to
-
08:41:37⏎The prospect of untreatable gonorrhoea has provoked alarm around the world, and there are no new classes of antibiotics in development.⏎In this week's Scrubbing Up column, Peter Green
-
ans with 10 limbs.⏎- Crab’s shell is a skeleton on the outside of its body.⏎- There are about 5,000 species of crabs.⏎- Crabs have claws on their front two legs.⏎- Cr
-
detector, the quicker you will be alerted to smoke or fire.⏎There are currently two different types of technologies used to detect smoke and fire. Photoelectric smoke detectors rely on an electric
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 first two, the first two, the first two, the first two, the first two, the first two. (This) the the the the the the the the the the the the the the the the the the the the the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
through all of it.⏎So what’s the key? Goals have to be actionable. That is
-
Buy second hand where possible and upcycle⏎- 9. Hang your clothes to dry rather than using a tumble dryer⏎- 10. Use all of your food leftovers⏎1. Reduce your meat intake⏎You don’t have to
-
keep on your desktop. And all the files get the “.mind” extension added to their filenames. Thus, the infection will not block you from accessing all of your files. However, it can do a lot of other things that can be
-
you’re internet dating a person from a further country, you will need to understand the customs that they are from so you can learn how to communicate properly with all of them. This can help you to enjoy your romance and avoid
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
- L4:1763 suppresses
- L3:31 excites
- L3:6 suppresses
- L3:763 suppresses
- L4:4 weakly suppresses
- L4:731 weakly excites
- L5:198 weakly excites unclear: weak "and" in lists
- L5:472 weakly excites
- L1:1796 weakly suppresses
- L0:784 weakly excites 'of' after 'think' or 'out'
- L3:624 weakly excites
- L3:1440 weakly excites
- L5:2838 weakly excites
- L2:1439 weakly suppresses
- L1:59 weakly suppresses 'the' after a verb or preposition
- L3:722 weakly suppresses
- L1:2567 weakly excites
- L5:38 weakly suppresses
- L2:2226 weakly suppresses
- L1:1430 weakly excites
- L4:1517 weakly excites
- L0:916 weakly suppresses
- L5:1372 weakly excites
- L5:1747 weakly suppresses
Feeds later layers
- L7:207 strongly suppresses
- L8:252 strongly suppresses 'of' after 'each' or 'one' or 'all'
- L14:280 strongly suppresses
- L16:1320 strongly suppresses
- L10:167 excites
- L8:636 excites
- L14:313 suppresses
- L17:985 excites
- L19:663 excites
- L13:975 excites
- L16:896 excites
- L17:1056 suppresses
- L19:1129 excites
- L19:1013 suppresses
- L19:1023 excites
- L21:274 suppresses
- L18:217 suppresses
- L15:557 suppresses
- L25:544 suppresses
- L21:310 suppresses
- L25:16 suppresses
- L7:173 excites
- L18:376 suppresses
- L18:602 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
- inion
- 成品
- result
- chte
- olarity
- biology
- RAM
- carp
- résultat
- clamation
Suppresses
- cade
- odesk
- 的一些
- 的一面
- dì
- 的时间
- pread
- 的缘
- 一些
- greSQL