the :// after http or https
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 fires on the '://' part of web addresses, right after 'http' or 'https'. It is weaker and sometimes negative on the later parts of a web address, such as the domain or the path. Turning it up gave off-topic text that did not clearly relate to links.
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.
-
:<|endoftext|>Copyright © University of Cambridge. All rights reserved.⏎'Aim High' printed from http://nrich.maths.org/⏎Why do this problem?⏎offers the opportunity for a
-
and we encourage you to use them however you feel will work best for your child and your own personal circumstances.⏎Check out the full range of plans here: https://www.edplace.com/blog/tag/learning_plans.⏎What are the EdPlace
-
ideal. False Dragonhead can tolerate both full sun and partial shade.<|endoftext|>Copyright © University of Cambridge. All rights reserved.⏎'The Solid' printed from http://nrich.maths.org/⏎is one of a series of problems designed to develop
-
up a new Python file and follow along:⏎import numpy as np⏎As you have to follow a few steps, I would suggest you go through this: https://www.thepythoncode.com/article/hide-secret-data-in-images-using-stegan
-
90/nu8020068⏎- Calcium/Vitamin D Requirements, Recommended Foods & Supplements.” National Osteoporosis Foundation, https://www.nof.org/p<|endoftext|>Growing tropical houseplants is an effective way to beaut
-
|Contact details of provider:|| Web page: http://www.tandfonline.com/VECE20|⏎|Order Information:||Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎
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: The problem is to find the number of possible... | of the first... This is a circle of size... This is a circle of size... the first... Based on the... of the first... This is a circle
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
With Circuit Diagram⏎Circuit Topology And Laws Applied Electricity⏎Solved Use Kirchhoff S Rules To Analyze The Circuit Diagram Chegg Com⏎Solved Problem 5 Consider The Circuit Diagram Shown Above
-
on to say… is the effect as lasting? Some believe not.⏎Taken from…. The Future of Education: Tablets vs. Textbooks http://mashable.com/2012/10/05/tablets-vs-textbooks
-
full article here: http://ourgame.mlblogs.com/2014/08/18/out-at-home/⏎- Related link: Click here to
-
Hakaayuʔ “Whitebread” (who told 19% of the stories) https://archive.org/details/traditionsofcadd00dorsrich⏎Explore more about the Caddo history and culture at Texas
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
- L11:2638 excites
- L4:2125 weakly suppresses
- L4:1267 weakly excites
- L8:665 weakly suppresses
- L1:2182 weakly suppresses
- L10:270 weakly suppresses
- L7:2495 weakly excites
- L12:1573 weakly suppresses
- L4:978 weakly suppresses
- L1:2627 weakly excites
- L12:2146 weakly suppresses
- L12:859 weakly excites
- L5:2165 weakly excites
- L4:1023 weakly suppresses
- L1:226 weakly excites
- L4:2596 weakly suppresses
- L10:1536 weakly suppresses
- L9:38 weakly suppresses
- L5:2402 weakly suppresses
- L8:1947 weakly excites
- L9:1961 weakly suppresses 'of' in titles and place names
- L3:2109 weakly suppresses
- L5:2584 weakly excites
- L6:2258 weakly suppresses
Feeds later layers
- L15:1978 suppresses
- L14:1432 suppresses
- L16:1603 suppresses
- L22:1040 weakly excites
- L14:2602 weakly suppresses
- L20:755 weakly suppresses
- L25:502 weakly excites
- L15:923 weakly suppresses
- L19:2376 weakly excites
- L23:844 weakly suppresses
- L17:2553 weakly suppresses
- L15:2148 weakly suppresses
- L25:638 weakly suppresses
- L23:2265 weakly suppresses
- L18:802 weakly suppresses line break after list items or paragraphs
- L22:957 weakly excites
- L15:1810 weakly suppresses
- L14:447 weakly excites
- L18:1162 weakly excites
- L22:2234 weakly excites
- L14:895 weakly suppresses
- L21:1918 weakly suppresses
- L23:544 weakly suppresses "Great" in proper place names
- L22:2987 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
- ieres
- ImageSharp
- Dou
- uong
- ึ
- uada
- 茬
- andle
- edar
- eros
Suppresses
- 深
- ющая
- ние
- 有钱
- Owners
- ashes
- mente
- 很快
- ...')⏎
- enville