'that' after 'added' or 'adding'
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 lights up on the word 'that' when it comes right after someone 'added' or 'adding' something, as in 'He added that...'. When it was turned up, the written text did not clearly change in a way we could link 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.
-
constant disappointment."⏎He added that technology now thrusts those alternate plans into our faces, which can make some people
-
shouldn't be eating or drinking.⏎Bener agreed it's difficult to know if consumption of sugary drinks has been falling much, adding that schools are only half the battle.⏎"Getting them<|endoftext|>The island of Guam is
-
access to information and communication technologies (ICTs) can mitigate many issues in SIDS and help to diversify their economies.⏎She added that innovations such as artificial intelligence, blockchain, drones, and mobile money were enabling progress
-
as it is today,” Bonde said, adding, "There were probably still people in the area and was probably a little bit marshy.”⏎He also added that the country's laws permit property owners to keep fossils discovered on their premises.⏎Talking
-
rely on focus groups of children to see what they like and what interests them. They have kid-testers participating in test sessions throughout the process. But she added that the team also spends a lot of time researching the developmental needs, social needs,
-
the butterflies feed or lay eggs,” said Rodrigues, adding that the Southern Birdwing (Troides minos), the biggest known butterfly in
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: I have a 3000000000 (A) style work of art. In this case, the title of the sculpture, the name of the work of art, is the same as the artist's self
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
said. “Still, you realize all rocket launches have some inherent risk, so, it’s just hold your breath and hope everything goes well.”Go Back to Top. Skip To: Start of Article.<|endoftext|>David Woods of Ohio State and Robin
-
poverty is represented mainly by achievements in education and health. The Human Development Index (HDI) that is regularly estimated by the United Nations Development Programme (UN
-
home for about 15 years. She struggles with some behavioral problems at the home and has difficulty communicating her needs to the staff.⏎Dr. Smith remarks, “I notice that Stephanie has never had a pap smear before, at least according to
-
in the past few years, said Deborah McCullough, a professor of entomology and forestry at Michigan State University. Based on her research, she says: “There is no reason for a landscape ash tree to die from emerald ash
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
- L5:1949 weakly excites
- L5:1003 weakly excites
- L3:1528 weakly excites
- L4:1484 weakly excites
- L2:2581 weakly excites
- L4:1349 weakly suppresses
- L1:1074 weakly excites
- L3:1679 weakly excites
- L5:881 weakly suppresses
- L3:1135 weakly excites
- L4:2058 weakly excites
- L1:722 weakly excites
- L2:1344 weakly suppresses
- L2:2064 weakly excites
- L4:2399 weakly suppresses
- L4:2898 weakly suppresses
- L2:410 weakly suppresses
- L3:2090 weakly suppresses
- L4:1104 weakly suppresses
- L5:316 weakly excites
- L2:253 weakly excites
- L4:1928 weakly excites unclear: personal names, mostly Indian
- L3:473 weakly excites
- L1:2895 weakly suppresses
Feeds later layers
- L9:1558 excites
- L7:983 excites
- L8:1929 excites
- L7:1028 excites
- L9:1900 excites
- L8:200 excites
- L9:2247 excites
- L9:2922 excites
- L8:2271 excites
- L10:653 suppresses
- L11:1876 excites
- L10:809 weakly excites
- L10:1517 weakly excites
- L14:2553 weakly suppresses
- L16:1757 weakly excites
- L7:1611 weakly excites
- L13:1569 weakly excites
- L8:1877 weakly suppresses
- L7:2815 weakly excites
- L21:3069 weakly suppresses
- L19:1807 weakly suppresses
- L11:2403 weakly excites
- L20:1666 weakly suppresses
- L18:464 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
- zac
- ovic
- 亲切
- 原谅
- DidLoad
- 话语
- 饼干
- breadcrumb
- requested
- olic
Suppresses
- SES
- comes
- SE
- endeavor
- TURE
- standing
- ?f
- YPE
- hir
- atem