'still' after 'but' in contrast
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. 13 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 'still' when it follows a 'but' or 'although' clause, as in 'but still a huge problem'. It marks a point where something is true despite a contrast. Turning it up sometimes gave text with 'still' and a 'but' twist, though not consistently.
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.
-
3% higher than the death rate for white people. In 2016, it was 16% higher. An improvement, yes — but still a huge discrepancy.⏎As with everything related to cancer, the factors contributing to these
-
increased from 8% to 84%, and that the number of reported head injuries and potentially serious head injuries decreased.⏎The percentage differences were minor, but still noteworthy.⏎A similar analysis was conducted by the NHS in the United Kingdom. They
-
of the living room, and several measures were made on this area. Although increased insulation improved the noise dampening capabilities of the embedded pipelines, the noise was still heard. Also the improvements could not explain the fact that the noise was also heard
-
oscillator producing a sine wave of about 37 kHz, which is then boosted to about 2400 volts. That’s pretty good, but nowhere near the 15 kilovolts needed for a self
-
31, and then slide steeply again until early 1933 before rebounding almost vertically for a few months, and then at a slower but still rapid pace.⏎Also See | The Recession Alphabet (Graphics)⏎The deep
-
set to give us plenty more of.⏎As expected, the hibernating (or desiccated) tardigrades fared better, but they’re still far from immortal. Their median lethal temperature was 82.7° C (
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: What is the current, which is the, which is the, which is the, which is the, which is the, which is the, which is the, which is the, which is the, which is the, which is the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
's disease, and how do you differentiate between regular memory loss and memory loss associated with Alzheimer's disease?⏎Answer: So, the memory changes of Alzheimer's disease represent a change for that person in
-
determine how newspaper articles and television reports describe and analyze the event. Use this...⏎6th - 12th English Language Arts CCSS: Adaptable⏎The State of the Union Address⏎Introduce your future voters to the annual State
-
such as product listings, were added. The new edition has also been changed by permitting the installation of decorative shrouds at the termination of a factory-built chimney<|endoftext|>When famous characters or people like Hercule Poirot
-
in which the student will answer two questions [50%].⏎Assessed Essay 1 will contribute 15% of the final mark for the module.<|endoftext|>Study the following vehicle parts for the type of vehicle you will be using during the
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
- L7:2736 weakly suppresses
- L12:2622 weakly excites
- L9:2777 weakly suppresses
- L12:2770 weakly suppresses
- L7:2832 weakly suppresses
- L0:637 weakly suppresses
- L5:1080 weakly suppresses
- L3:1223 weakly excites
- L9:512 weakly suppresses
- L11:482 weakly excites
- L7:2018 weakly suppresses
- L9:962 weakly excites
- L11:2749 weakly excites
- L8:3020 weakly excites
- L5:1512 weakly excites
- L10:2371 weakly suppresses
- L6:2015 weakly excites
- L4:1663 weakly excites
- L7:809 weakly excites
- L7:126 weakly excites
- L2:2380 weakly suppresses
- L7:2738 weakly suppresses
- L5:2110 weakly excites
- L4:1183 weakly suppresses
Feeds later layers
- L14:1123 excites
- L16:2778 excites
- L14:1285 excites
- L14:1361 excites
- L14:2809 excites
- L14:2171 weakly suppresses
- L18:2340 weakly excites
- L14:536 weakly suppresses
- L21:864 weakly suppresses 'concentrated' and comparative adverbs
- L15:1422 weakly excites
- L14:2034 weakly excites
- L14:1903 weakly excites
- L14:2860 weakly suppresses
- L15:250 weakly excites
- L15:649 weakly excites
- L16:2748 weakly suppresses
- L14:1131 weakly suppresses
- L14:1057 weakly excites
- L23:628 weakly excites
- L20:1268 weakly suppresses
- L14:2712 weakly suppresses
- L19:1454 weakly excites
- L18:2883 weakly excites
- L15:2078 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
- fires
- ernes
- že
- ILED
- ируем
- enes
- 层层
- enie
- anje
- grease
Suppresses
- 湛
- views
- .addAttribute
- 市教育局
- ToShow
- /o
- Admission
- !/
- chestra
- views