'and' joining two abstract nouns
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. 26 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 'and' when it links two nouns or ideas, such as 'self-awareness and physical understanding' or 'support and motivation'. Turned up, the model repeated words like 'always'; turned down, it repeated 'follow' and 'work'.
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.
-
make them citizens of mathematics. Through collegial exchange of free imagination and logical argument, they gain admiration for each other and self-esteem as the harmonious forms they are studying emerge.⏎The class leader simply
-
se psychologist and counselors at a group called the Bethesda Counselling Centre are helping Kaswera — and others who lost loved ones to Ebola — to find solace and closure through alternative burial rituals.⏎With a team of four counselors, psychologist Noé
-
"In the future, it will no longer be possible to make it through one's entire career just by graduating today. We all need to continually improve our qualifications and learn new things," says
-
psychophysical way to learn about ourselves through our personal movement choices. By exploring everyday movements, work and performance with the Alexander Technique, you can increase self-awareness and physical understanding. This understanding can lead to relief from pain, stress and tension,
-
, you will be leaving the KidsHealth site.⏎Sickle Cell Disease Association of America⏎This group provides education, advocacy, and other initiatives to promote awareness of and support for sickle cell disease programs.⏎National Institute of Neurological Disorders and Stroke
-
ocial therapy or a combination of both for proper treatment of the disorder.<|endoftext|>Goals and resolutions⏎Goals are an essential part of life. They provide us with direction and motivation to work towards something greater than ourselves. Goals give us a sense of purpose
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: follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow follow
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. There is considerable variation in the canine population regarding the types of responses that are generated by fear-provoking stimuli. Some dogs require a very strong stimulus to elicit fear, while others become extremely anxious in response to mild stimuli or any auditory
-
how he prefers to classify recessions.⏎Inherently, what is the main problem with classifying recoveries into alphabet shapes? What sort of complexities does it elide over?⏎Quite simply, if you actually look at the broad measures of aggregate
-
OC’s represent a fundamental change in the way students are confronted with material and how they engage in it. The essence of a MOOC is to activate, to elicit interest, action and debate about the content between the students themselves. Short videos
-
It also shows, from one lung cancer to another, the cancer stem cells are not the same."⏎Cancer stem cells are a subset of cancer cells believed to elude conventional treatments and eventually regenerate a tumor. Experimentally, they show up as
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
- L24:1080 strongly suppresses
- L25:364 strongly suppresses
- L24:298 strongly excites 'would' in hypothetical or past statements
- L22:627 strongly suppresses
- L21:853 strongly excites 'can' after 'and' or a noun subject
- L24:314 strongly excites
- L19:3044 strongly suppresses
- L23:505 strongly suppresses
- L21:1708 strongly excites
- L21:1657 strongly suppresses
- L23:373 strongly suppresses 'to' after a noun or adjective
- L20:636 strongly suppresses
- L23:257 strongly suppresses
- L25:125 suppresses
- L25:2260 excites
- L25:906 suppresses
- L17:1297 excites
- L25:418 excites
- L25:361 suppresses
- L21:322 excites
- L25:104 suppresses
- L19:858 suppresses
- L20:2802 suppresses
- L20:349 excites
Feeds later layers
- L27:515 strongly suppresses
- L27:401 strongly suppresses
- L27:247 strongly suppresses
- L27:201 strongly suppresses
- L27:488 suppresses
- L27:880 suppresses
- L27:122 suppresses
- L27:963 suppresses
- L27:577 suppresses
- L27:559 suppresses
- L27:900 suppresses
- L27:645 suppresses
- L27:434 suppresses
- L27:883 suppresses
- L27:801 weakly suppresses
- L27:249 weakly excites
- L27:520 weakly suppresses
- L27:1493 weakly suppresses
- L27:1129 weakly suppresses
- L27:920 weakly suppresses
- L27:1126 weakly suppresses
- L27:1059 weakly suppresses
- L27:273 weakly suppresses
- L27:1283 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
- eil
- 的方向
- BOOK
- 금
- ion
- reak
- edTextBox
- OptionPane
- 点了
- 子
Suppresses
- follow
- use
- resolve
- hold
- look
- offer
- reach
- tend
- take
- carry