'seen' after 'never' or 'rarely'
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. 0 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 responds to the word 'seen' when it comes after words like 'never', 'rarely', or 'last', as in 'had never seen'. Turning it up did not produce a clear, consistent change in the writing.
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.
-
of its western relative, the Ivory-Billed Woodpecker. The bird was last seen in 1944, and recently has been declared extinct.⏎A kay
-
To be honest, those fruits look like the breasts of a cow. I have never seen this plant in any other country, yet it is hard to imagine something so flashy
-
focused on to overcome many of the more frequently seen injuries by orthopaedic therapists. It is these similarities which have encouraged many therapists
-
Matt 2008?” video is a great example of the impact of music on images and changes how we look at it.⏎In case you haven’t seen this,
-
, to thin the herds. So far, it is working.⏎Contrary to my story, wolves don’t normally attack people. They’re shy animals rarely seen by visitors.⏎Until releases this week from Science Advances, it was believed there were
-
spy lots of different kinds of leaves lying on the ground. Some were leaves I had never seen. Some were still green, while others were changing to their autumn colors.⏎Have
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 problem of the old man. I have been thinking about it for a long time. It's a problem. It's a problem. It's a problem. It's a problem. It's a problem. It's a problem.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
(reigned 1187–56 bce) who defended his
-
(So do we, which is why we laugh at this monk's struggles.)⏎Here
-
(GNBDK)⏎Never do to anyone else anything that you would not want someone
-
(p. 136) and recognized by Hans Reichelt (pp.
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
None: this is the first layer.
Feeds later layers
- L9:785 weakly excites
- L4:296 weakly suppresses
- L18:611 weakly suppresses
- L4:18 weakly suppresses
- L17:466 weakly suppresses
- L1:110 weakly suppresses
- L1:1210 weakly suppresses
- L1:2147 weakly suppresses
- L6:226 weakly excites
- L12:2174 weakly suppresses mixed, web boilerplate phrases
- L19:194 weakly suppresses
- L18:1078 weakly suppresses
- L1:2677 weakly suppresses
- L3:574 weakly excites
- L17:2489 weakly excites
- L7:2660 weakly suppresses
- L7:1331 weakly suppresses
- L18:1329 weakly suppresses
- L4:1701 weakly suppresses
- L13:499 weakly suppresses
- L1:2461 weakly suppresses
- L21:518 weakly suppresses
- L21:486 weakly suppresses
- L14:2321 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
- elves
- ext
- 出口
- 复苏
- iff
- 意图
- uard
- led
- umble
- spd
Suppresses
- verage
- -initialized
- deltas
- те
- aye
- ólogo
- Paginator
- اما
- NavBar
- 持有的