spot right before a number
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 reacts to the word or space just before a number is about to appear, like 'born in' followed by a year, or 'There were' followed by a count. Turning it up did not produce a clear theme in the replies.
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.
-
(6-8) Show all Grade Level⏎Resource Type: Activities⏎Results 1 - 7 of 7 matches⏎From Grid
-
Shoguns.⏎It is also known that there was once an ancient civilisation in the area pre<|endoftext|>- Herbert A. Simon was an American scientist born in 1916.⏎- He is most famous for his work in the field
-
during the final manned flight of the Skylab space station in 1973 and 1974.⏎He was born January 23, 1930, in Okemah, Oklahoma. He received a Bachelor
-
and 1.79% from two or more races. Hispanic or Latino of any race were 0.89% of the population.⏎There were 46 households out of which 28.3% had children under the
-
jing Xiaozhuang University, Nanjing, Jiangsu 211171, China⏎*Authors to whom correspondence should be addressed.⏎Received: 2020-8-4 / Accepted: 202
-
created with an extra letter added to rifts:⏎24 words found using the letters in "Rifts"⏎5 Letter Words⏎4 Letter Words⏎3 Letter Words⏎2 Letter Words<|endoftext|>Marx and Engels' work covers a wide
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: and I want to know about this, and also to know about this, and also to know about this, and also to know about this, and also to know about this, and also to know about this, and also to know about
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
in Education (eJSiE):⏎2, Article 1.⏎Available at: http://epublications.bond.edu.au/ejsie/vol8/iss2/1<|endoftext|>Some may argue that sleeping is one of the best parts
-
last edited on: 2008-09-09 15:47:35⏎Did you find this article helpful? Yes No⏎How can we improve this article?<|endoftext|>Java tutorials
-
adapt. [Awareness of climate change] is<|endoftext|>- Created by: Luke Simons⏎- Created on: 06-04-12 19:17⏎1. Newton's model of the universe made four
-
here. There are several questions in this exercise. Our subject expert teachers have solved each of these questions in a step by step manner. All these solutions are beneficial for the students of class 10 appearing for the board exams. Students can easily
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
- L12:2530 weakly suppresses
- L6:940 weakly suppresses
- L9:857 weakly excites
- L11:2083 weakly excites
- L7:1558 weakly excites
- L8:424 weakly suppresses
- L12:1045 weakly suppresses
- L9:2396 weakly suppresses
- L11:546 weakly suppresses
- L8:229 weakly excites
- L5:496 weakly suppresses
- L12:2177 weakly suppresses
- L6:725 weakly excites
- L4:2456 weakly excites
- L8:2893 weakly suppresses
- L10:1553 weakly suppresses
- L10:2598 weakly excites
- L9:50 weakly suppresses
- L11:1022 weakly excites
- L8:339 weakly excites
- L8:1761 weakly excites
- L7:2272 weakly suppresses
- L10:2939 weakly suppresses
- L7:2464 weakly excites
Feeds later layers
- L17:2529 suppresses
- L14:1002 weakly excites
- L17:2535 weakly excites
- L25:3049 weakly suppresses
- L16:1920 weakly excites
- L15:339 weakly excites
- L20:631 weakly suppresses
- L17:2704 weakly suppresses
- L19:2601 weakly suppresses
- L15:234 weakly suppresses
- L18:2581 weakly suppresses
- L14:2788 weakly suppresses
- L17:1078 weakly excites
- L14:2439 weakly excites
- L27:1205 weakly excites
- L18:266 weakly excites
- L16:2568 weakly excites
- L14:2298 weakly suppresses
- L14:229 weakly suppresses
- L14:1051 weakly excites
- L16:1380 weakly excites
- L14:2559 weakly suppresses
- L27:376 weakly suppresses
- L16:2337 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
- iments
- agram
- VES
- getService
- ungle
- شف
- LAS
- 工业大学
- uffers
- visualization
Suppresses
- 带
- 甚至是
- or
- 或者
- 甚至
- 存在
- çekil
- chap
- 自理
- 印