'Aut' starts and colon prediction
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. 27 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
It fires on the opening piece 'Aut' in words like 'autism' and 'autotransformer', and on some other word starts. Its output pushes the model toward writing a colon, and turning it up made the model write long strings of colons.
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.
-
they were able to share stories and discuss challenges. Although the ages of the children and levels of severity vary, members of the group have found much common ground.⏎Autism spectrum disorder
-
condition]⏎wiki: In the past, syndromes were often named after the physician or scientist who identified and described the condition in an initial publication⏎ex. 唐氏
-
become a foolhardy risk of life as well as a waste of time and energy.⏎So the strategy changes: evacuate the vulnerable, the impossible to defend. Defend that which can be defended, wait for the fire to burn itself out and
-
functional parts.⏎- 1 Power transformers⏎- 1.1 Laminated core⏎- 1.2 Toroidal⏎- 1.3 Autotransformer⏎- 1.4 Variable autotransformer⏎-
-
the production of silicon. The activity is an investigation of silicon: the sources, uses, properties, importance in the fields of photovoltaics (solar cells/renewable energy) and integrated circuits industries, and, to a limited extent,
-
?⏎Suddenly, you would be a mess. You’d ask for help from your parents, and would likely be embarrassed that you had to ask them for help.⏎Autistic individuals are often in this situation. Autistic
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: sometimes it seems that way.",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".",".","
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and criteria as we use for PAA and CAF.⏎Through these three programs each<|endoftext|>发布时间:2017-03-27 来源:新
-
.⏎2 پژوهشگر دوره تخصصی دکتری معماری منظر، دانشکده معماری و شهرسازی، دانش
-
there is the risk of a major collapse in agricultural production in poor countries. This will increase malnutrition already at elevated levels in these countries.⏎Sr. Bénédicte Kikwengi, SNDdeN works with her students at
-
very low in calories, which makes<|endoftext|>A Study on Chinese Auxiliary Verbs for Thai Learners Based on the Comparative Analysis⏎- 人文学院-已发表论文 ⏎“助动词”在汉语中具有较大的作用,使用率高
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
- L26:345 strongly excites
- L26:1329 strongly excites
- L25:737 strongly excites
- L26:2215 suppresses
- L24:1896 excites
- L24:500 excites
- L19:560 excites end of document, and 'it turned out'
- L26:122 suppresses
- L25:502 suppresses
- L26:582 excites
- L24:853 suppresses
- L21:165 excites
- L20:2106 excites
- L26:823 excites
- L23:236 excites
- L20:54 excites
- L14:989 weakly excites
- L24:2971 weakly suppresses
- L21:616 weakly excites
- L20:1313 weakly excites
- L3:2524 weakly excites
- L25:4 weakly excites
- L25:1050 weakly excites
- L21:2823 weakly excites
Feeds later layers
None: this is the last layer.
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
- :
- :
- :⏎
- :⏎⏎
- :⏎
- :⏎⏎
- : ⏎
- :")⏎
- :"
- :");⏎
Suppresses
- �
- ってしまいます
- є
- uchen
- fortunate
- .","
- enerative
- afone
- .','
- pleaded