'the' right before 'subject' or 'source'
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. 21 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 reacts to the word 'the' in sentences like 'has been the subject of study' or 'became the source of conflict', where a word like 'subject' is expected next. When it was turned up, the model kept producing 'the subject of the focus' again and again.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
means, and these attempts have a high scientific interest on account of the mystery which surrounds the natural origin of this remarkable mineral. Its physical and chemical properties have been the subject of much study
-
accessed here. You can also read the full article published in Transactions of the Institute of British Geographers here.⏎Ethnic diversity in English city neighbourhoods has been the subject of longstanding public interest. While some have argued that diversity is associated with lower
-
under a Creative Commons Attribution 4.0 International License.<|endoftext|>How do you explain how babies are made to an 8-year-old?⏎This topic has been the subject of countless jokes over the
-
practices. But at the same time, there are enough differences to make the two religions and cultures oppose each other greatly. Even some similarities between the two have been the source of conflict for thousands of years.⏎Both the Jewish and Muslim faith believe in
-
. There are over a hundred lakes and ponds in the mire, many quite small. Only a few have names; these are medium-sized lakes which have been the subject of several studies.⏎On the map below several lakes are named. Using the
-
and covering the period from 1319 to 1367. The book's stories had great popularity in the nineteenth century, when they became the subject of musha-e, prints depicting warriors, drawn by Kuniyoshi
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: this is a 8-digit a,3,3,3,3,3,3,2,3,3,2,2,3,3,3,2,3,3,2,2,3,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Never Go Swimming Alone⏎You should continuously swim with a mate unless you're swimming in a swimming pool or a lake.⏎Remember that the lifeguard isn't your "swim buddy"; when on duty, they have a lot of people to
-
Dehydration can make you more vulnerable to illness and infections, therefore, it is important that you down plenty of non-alcoholic fluids. And if water isn’t your thing, juice, tea, and other beverages
-
washed away oils.⏎This can make your ferret smell worse.⏎If you are looking for a hypoallergenic pet, then a ferret may not be your best choice. However, if you don’t have any sensitivities to particular smells
-
(Gardenia spp.) are recommended for growth in U.S. Department of Agriculture plant hardiness zones 8 and above, cool temperatures are not the plant’s only enemy. The plant also requires humidity to look its best, which is why
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
- L19:2531 excites
- L20:2247 suppresses
- L18:2949 excites
- L20:1540 suppresses
- L8:2519 excites
- L15:1668 suppresses
- L15:2692 excites 'to' after a noun, before a purpose verb
- L18:2232 excites
- L18:780 suppresses
- L17:1310 suppresses
- L17:2985 excites
- L17:1262 excites
- L19:2132 excites
- L20:1253 suppresses
- L16:2038 suppresses
- L5:1395 excites
- L5:2141 excites
- L18:1502 suppresses
- L17:2756 suppresses
- L20:2026 excites
- L19:1910 weakly suppresses
- L16:2091 weakly suppresses
- L19:820 weakly excites
- L15:2257 weakly suppresses unclear: word stems in technical text
Feeds later layers
- L22:2708 excites
- L24:2468 excites
- L27:2480 excites
- L25:2948 excites
- L23:916 excites
- L27:127 suppresses
- L25:2215 excites
- L27:399 weakly excites
- L23:1486 weakly excites
- L23:2966 weakly excites
- L23:2711 weakly excites
- L25:2040 weakly excites
- L26:507 weakly excites
- L23:2838 weakly excites
- L22:1336 weakly excites
- L27:1039 weakly excites
- L22:438 weakly excites
- L27:1376 weakly excites
- L23:94 weakly suppresses
- L27:1292 weakly suppresses
- L23:556 weakly suppresses
- L23:748 weakly excites
- L27:1327 weakly excites
- L23:2951 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
- subject
- basis
- subjects
- 对象
- source
- focus
- subject
- Subject
- object
- basis
Suppresses
- STER
- ewise
- ApplicationBuilder
- asz
- 行った
- Dungeons
- asti
- ingly
- 记者采访
- Stern