"for" after "hard" or "difficult"
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. 2 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 "for" in phrases like "hard for", "difficult for" or "easier for", right before the person affected is named. Turning it up did not produce an obvious change in the text.
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.
-
radiation, and other agents, are published in a Special Report in the August issue of The Lancet Oncology.⏎Source: Medical News Today<|endoftext|>It’s hard for even the most rigorous of scientists to argue with an elegant experimental design and a pretty
-
the impact of completely remaking the lives of many working people after the war. The US entered a time of great prosperity after the war, making it much easier for working people to make a good living. In this sense, the war was a
-
. This makes it difficult for businesses to forecast the cost of fuel and can cast doubts over the future of their
-
the soil is fit for planting, probably past the best time for sowing.⏎The loss of draft animals will make it hard for farmers to<|endoftext|>Calò is a prime example of the merging of language and culture
-
the depreciating mentality of Columbus and his hopes. As each voyage is unsuccessful in producing Oriental splendors or in establishing relations with the Great Kahn, it becomes harder for Columbus to persuade others to support his missions
-
fear of missing out'⏎CHARLESTON, W.Va. -- As the constant need for communication is becoming the social norm, it's increasingly difficult for people
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 been working on it. I have been writing it. I have been thinking about it. I have been writing it again. I have been thinking about it again. I have been writing it again. I have been thinking about it
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the gasoline-powered fleet.⏎That’s according to the new data published from the University of Chicago, the University of California, Davis, and the University of
-
agent(s) by a route other than IV; (PO, PR, Intranasal, IM) and appropriate monitoring.⏎Regional Anesthesia: A term used for local anesthesia.⏎Removal of the tip of a tooth root.<|endoftext|>The Adams
-
mice into eight separate lines – four lines bred for high levels of daily running, with the remaining four used as controls – and measured
-
–5, 6–8, 9–12⏎Important events during the presidency of Jimmy Carter.⏎President Jimmy Carter agreed to answer seven questions selected from several hundred posted by Scholastic students.⏎A time line showing the events in
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
- L1:1963 weakly excites
- L1:2500 weakly excites
- L0:1244 weakly suppresses
- L1:1804 weakly suppresses
- L0:1543 weakly excites
- L0:1778 weakly suppresses
- L1:1230 weakly excites
- L1:2330 weakly suppresses
- L0:1447 weakly excites
- L0:2989 weakly suppresses
- L1:1892 weakly excites
- L0:2922 weakly suppresses
- L0:2765 weakly excites
- L1:2720 weakly suppresses
- L0:2301 weakly excites
- L0:931 weakly excites
- L0:897 weakly suppresses
- L0:84 weakly suppresses
- L0:1369 weakly suppresses
- L0:376 weakly excites
- L0:3 weakly suppresses
- L1:1960 weakly excites
- L1:92 weakly excites
- L0:1852 weakly suppresses
Feeds later layers
- L12:2535 excites
- L7:304 excites
- L23:1062 excites
- L8:2680 weakly excites
- L25:343 weakly excites
- L14:2150 weakly excites
- L5:2365 weakly suppresses
- L20:2180 weakly excites
- L20:2161 weakly excites
- L10:2426 weakly excites
- L20:2229 weakly excites
- L25:1639 weakly excites
- L15:2843 weakly excites
- L9:654 weakly excites
- L23:2933 weakly excites
- L22:94 weakly suppresses
- L18:2120 weakly excites
- L14:1169 weakly excites
- L13:1162 weakly excites
- L23:436 weakly excites
- L25:1742 weakly excites
- L6:1547 weakly suppresses
- L11:2227 weakly suppresses
- L20:954 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
- ocht
- rometer
- rieb
- translator
- orque
- correspondent
- :x
- endant
- orary
- criança
Suppresses
- led
- Citation
- 免责声明
- amation
- tà
- ."""⏎⏎
- .gdx
- ['$
- glGen
- Sundays