evaluative words: good, bad, positive
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. 10 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 words that judge something as good or bad, such as 'favorable', 'good terms', 'bad rap' and 'positively'. Turning it up did not produce a clear change in what the model wrote about.
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.
-
is too jaded by his ambition.⏎Although this essay is historically accurate it lacks important details, which might paint a different view of Columbus. Boorstin writes favorable of Columbus and depicts him as a heroic and determined figure who helped shape history,
-
feuding families eventually stopped but only because Romeo and Juliet died. This was mainly because of the Friar.⏎The Capulets and Montagues were not on good terms with each other. Unfortunately though, Romeo and Juliet fell in love with each
-
age or time. Teach them that a good person chooses to think good thoughts, not to say profaned words, do good deeds, go to good places
-
of heart disease.⏎Carbohydrates have gotten a bad rap lately, influenced by various fad diets popular today. As a result,
-
the studies.⏎This makes one wonder if the collagen used in<|endoftext|>The passive voice in Spanish is one of those strange grammatical expressions that strikes fear into the best of us. The truth of the matter is that one of the challenges of learning a
-
, Irish is the expression of a cherished culture, but its close association with nationalism and republicanism means that Protestants and unionists rarely see it in a positive light. History reveals that this was not always the case. For centuries, Protestants
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: $ S_n = d_k h_k l_k h_k l_0 h_k l_0 h_k l_0 h_0 h_k l_0 h_k l_0 h_0 h_0 h_k l_
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
was written amidst the 1800’s. Hawthorne was a famous American author during that time frame. He is a relative to a judge from the Salem Witch Trials, which was his Great-Great Grandfather John Hathorne. Hath
-
life forces and increase energy and strength. Ruby is also known as the stone of courage, and legends record that a person possessing a ruby can walk through life without fear, evil or experiencing misfortune.⏎Throughout gemstone history, spinel and garn
-
more than one anxiety disorder. Sometimes anxiety results from a medical condition that needs treatment.⏎Common symptoms of anxiety would be increased
-
over on a stool in the 15th Evacuation Hospital in Sicily. A husband and father from South Bend, Indiana, had fought bravely for eight months through the Tunisian and Sicilian campaigns. Within the last two weeks
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
- L7:2333 excites
- L7:1110 suppresses
- L6:2241 excites
- L7:1674 suppresses
- L8:2784 excites
- L6:1821 suppresses
- L6:2847 suppresses
- L8:1978 suppresses
- L8:2637 suppresses
- L8:2279 excites
- L1:1263 suppresses
- L9:667 suppresses
- L7:1586 suppresses
- L7:1839 weakly excites
- L8:2899 weakly suppresses
- L8:1143 weakly suppresses
- L9:2963 weakly excites
- L6:222 weakly suppresses
- L7:1769 weakly suppresses
- L9:1623 weakly suppresses
- L9:258 weakly excites
- L9:2014 weakly suppresses
- L8:1071 weakly excites
- L4:1476 weakly suppresses
Feeds later layers
- L11:2178 excites
- L20:1384 excites
- L16:2617 excites
- L13:2350 suppresses
- L14:2623 excites
- L12:1749 excites
- L24:374 excites
- L19:2952 excites
- L18:2051 excites
- L13:1981 excites
- L15:2970 weakly excites
- L16:2756 weakly excites
- L12:2764 weakly excites
- L12:1886 weakly excites
- L23:1592 weakly excites
- L14:1286 weakly excites
- L21:1689 weakly excites
- L18:2187 weakly excites
- L21:2186 weakly excites
- L15:591 weakly suppresses
- L11:971 weakly excites
- L13:2414 weakly excites
- L11:2123 weakly excites
- L18:1698 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
- -negative
- 正面
- positive
- negative
- Negative
- 恶劣
- Positive
- adverse
- negative
- 负面
Suppresses
- elts
- @",
- \base
- taller
- 門
- @}
- byname
- recre
- bran
- towers