verbs of finding out or thinking
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. 12 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 verbs like 'know', 'believe', 'find out', 'think' and 'track' when they lead into a clause or object. Its weaker negative side fires on words like 'about' and 'of' after verbs of talking. The steering results did not show a clear theme when turned up.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
a topic or thesis into a logical order. Outlines arrange major topics, subtopics, and supporting details. Writers use outlines when writing their papers in order to know which topic to cover in what order.⏎What is outline
-
More Info... by email⏎Consumer ethnocentrism is a psychological concept that refers to individuals who believe that their country's products are superior to those of other countries. This concept also
-
rial polyps. The first test is a transvaginal ultrasound. This is done vaginally using a wand-like device that emits sound waves to create a picture of the uterus. If the patient has any polyps, the ultrasound will detect
-
know how EGT1 helps roots to overcome gravity. For this, the researchers looked at the genes disrupted in the EGT1 mutant. This, it turned out, were genes for producing parts of the cell wall. This disruption translated into less
-
, 1975⏎- The Exxon Valdez, 1994⏎Using the Internet, in a limited period of time, students find out:⏎1. What
-
may be due in part to the hormonal upheavals that occur during the teen years. But be careful – it’s easy for adults to take this information and think “it’s just hormones” and therefore think the depression does not need to be
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: I have a 3D model of a 3D model of a 3D model of a 3D model of a 3D model of a 3D model of a 3D model of a 3D
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the letters we use so often. The earliest forms of writing evolved because members of ancient civilizations needed more efficient ways to express themselves. Listen to this story to hear about the origins of individual letters, as well as learn about the connection between shape and
-
ne Davis, works as a Postdoctoral research fellow in the Plant and crop science department at the University of Nottingham.⏎We asked her about the role of women and girls in science and how the situation has changed over the
-
not niche. They are not elective or historical. They reach right into those classrooms, affecting every single student. Students we worked with talked of being cat-called, harassed, objectified, condescended to and, most
-
to further engage conversation.⏎Vocabulary lists are used in almost every classroom and rarely offer any interactive engagement for students. Join us as we talk about Thinglink, a website that allows students to create<|endoftext|>Pulley Reef (
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
- L11:1774 excites
- L10:653 excites
- L11:2537 excites
- L10:2275 suppresses
- L9:2247 suppresses
- L11:1469 excites
- L9:2172 excites
- L9:761 excites 'sure' in 'are you sure' phrasing
- L9:1512 suppresses
- L10:1236 weakly excites
- L9:512 weakly excites
- L10:2208 weakly excites
- L11:1036 weakly excites
- L9:2377 weakly suppresses
- L11:1715 weakly suppresses
- L5:2877 weakly excites
- L10:551 weakly suppresses
- L8:2209 weakly excites
- L10:1941 weakly suppresses
- L11:783 weakly excites
- L11:1292 weakly excites
- L10:2032 weakly excites
- L4:1366 weakly excites
- L4:666 weakly suppresses the digit 0 in dates
Feeds later layers
- L14:1091 excites
- L13:743 excites
- L16:1019 suppresses
- L17:1296 excites
- L16:1344 excites
- L17:2038 suppresses
- L14:1385 excites
- L16:2210 weakly suppresses
- L20:27 weakly suppresses
- L14:421 weakly suppresses
- L14:2651 weakly excites
- L13:1933 weakly suppresses
- L21:655 weakly suppresses
- L13:1851 weakly excites
- L15:598 weakly excites
- L18:317 weakly suppresses
- L17:2784 weakly excites
- L19:290 weakly suppresses
- L18:1027 weakly suppresses
- L16:2184 weakly excites
- L15:2105 weakly suppresses
- L14:1894 weakly excites
- L13:1046 weakly excites 'in' or 'on' ending a phrase
- L16:457 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
- afen
- específico
- -packages
- 津
- ้อน
- ython
- 某一
- essim
- CONSEQUENTIAL
- cpy
Suppresses
- ARING
- fore
- ffect
- ieve
- .fm
- whel
- mere
- 임
- 广场
- LETE