decimal points inside numbers
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 lights up on the dot in numbers like 3.0, 2.5, 1.25 or 10.4, and on version or citation numbers. When it was turned up, the model fell into repeating strings of code-like symbols, which tells us little about the neuron itself.
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.
-
List of Users.⏎Sometimes, it’s also handy to define a list of multiple Entities and Structures.⏎Until 184.108.40.2
-
for forest health, natural resources and silviculture images operated by The Bugwood Network at the University of Georgia and the USDA Forest Service., CC BY 3.
-
- Amount of protein: Experts recommend adults over age 65 consume .45 to .68 grams of protein per pound of body weight daily (1.0-1.5 gm/kg/d), depending on
-
the FN FAL, the HK G3 and the AK-47.⏎A magazine-fed, select-fire rifle firing the .308 Winchester/7.62x51mm NATO was, at the time, seen as the
-
on this website is licensed under a Creative Commons Attribution 2.5 Australia Licence together with any terms, conditions and exclusions as set out in
-
all the texts on this journal belong to the respective authors without restrictions.⏎This journal is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (full legal code).⏎See also our Open Access Policy.
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 know that, well, two of the top top! top! top! top! top! top! top! top! top! top! top! top! top! top! top! top! top! top! top!
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
return to a shielded base.⏎To continue reading this article, subscribe to receive access to all of newscientist.com, including 20 years of archive content.<|endoftext|>Fly Created To Study How A Normal Cell Turns Cancerous⏎The study
-
will review what you’ve submitted and determine whether to revise the article.Join Britannica's Publishing Partner Program and our community of experts to gain a global audience for your work!⏎One of the youngest of the world’s major cities, Johannesburg was founded in
-
explaining how they use data to make predictions and offering a comprehensive understanding of this fascinating topic. Get ready to explore the world of decision trees and discover how they make predictions that are accurate, reliable, and effective.⏎II. What are Decision Trees?⏎
-
and considering dental treatments might aid in detecting and preventing oral diseases before they become more serious. Supreme Dental CT helps you to keep a beautiful smile and improve your overall health by making these practices a habit.<|endoftext|>Teen Attitude: What is normal?
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
- L20:44 strongly excites
- L18:2322 strongly suppresses end-of-text token after a finished passage
- L17:1088 strongly excites
- L17:658 strongly suppresses
- L16:235 strongly suppresses
- L25:808 strongly suppresses
- L15:774 strongly excites
- L18:926 excites
- L17:1523 suppresses
- L17:68 suppresses
- L17:1029 excites
- L25:118 excites
- L16:2510 excites
- L15:2675 excites
- L16:3064 excites
- L15:101 excites
- L17:2723 excites
- L17:3036 excites end of document boundary
- L18:791 suppresses
- L15:917 suppresses
- L17:2848 suppresses
- L14:667 suppresses
- L24:36 excites
- L17:450 suppresses
Feeds later layers
- L27:725 strongly excites
- L27:200 suppresses
- L27:1078 weakly excites
- L27:136 weakly excites
- L27:114 weakly suppresses
- L27:502 weakly excites
- L27:531 weakly excites
- L27:220 weakly suppresses end-of-text token between documents
- L27:1899 weakly excites
- L27:1530 weakly excites
- L27:170 weakly suppresses unclear: spaces in mixed-language text
- L27:207 weakly suppresses
- L27:2795 weakly excites
- L27:697 weakly excites
- L27:690 weakly suppresses
- L27:1554 weakly suppresses
- L27:1255 weakly suppresses
- L27:1300 weakly excites
- L27:313 weakly suppresses
- L27:1909 weakly excites
- L27:2964 weakly excites
- L27:1399 weakly excites
- L27:126 weakly suppresses
- L27:1089 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
- <|endoftext|>
- 返回搜狐
- […]⏎⏎
- ↩
- ')}}</
- ⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎
- ⏎ ⏎ ⏎ ⏎
- }}</
- ']}}</
- [...]⏎⏎
Suppresses
- _',
- odont
- IVITY
- Ń
- opsis
- üstü
- полу
- .minecraft
- adata
- ulos