word before a polite or emotional phrase
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. 9 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 fires on small connecting words that come right before a set phrase, like 'sorry for', 'thanks for', 'proud to' and 'take heart'. It seems to care about the words leading into an apology, thanks or feeling. Turning it up did not make the writing emotional, so the pattern is hard to confirm from the outputs.
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.
-
code there is easily understandable though...⏎January 10th, 2004 01:32 AM⏎Tim_axe, sorry to be pedantic, but if you look at his code, x would never be zero
-
America's working men and women by setting and enforcing standards, and providing training, education and assistance. For more information, visit www.osha.gov.<|endoftext|>End of the year math review for 2nd grade can be FUN! 2nd
-
pressure. As the resulting jets of gas shoot backward, the engine and its craft are thrust forward. However, showing you is much cooler than explaining it.⏎Thanks for Subscribing!⏎You'll need to confirm your email, so please check your
-
for college and career.⏎The statistics are sobering, but take heart! There are steps you can take to help prepare your teens for college.⏎College
-
elacanth specimen was captured by fishermen in 1938. This inflicted a terrible disappointment on evolutionists. James Leonard Brierley Smith, an instructor in the Rhodes
-
to their final size on the map. I'm afraid I was unable to find the original author of the tutorial I used to create these planets, so I apologise for not being able to offer credit where it is due.⏎Make a new image with
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 want to make a point that: (1) the value of the expression is that the number of the expression is that the value of the expression is that the number of the expression is that the value of the expression is that the number
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
partners, explore the possible health and economic impacts of using pricing strategies to promote consumption of healthy foods and beverages and discourage consumption of unhealthy products.<|endoftext|>ADS BY GOOGLE⏎In grammarFrom a grammatical point of view, superlative is a type
-
<|endoftext|>CONSUMER ADVISORY: Motorists Should 'Share the Road' With Motorcyclists⏎Friday, May
-
designer, one of the greatest names in French fashion in the 20th century. In 1985, Caroline Rennolds Milbank wrote, "The most consistently celebrated and influential designer of the past twenty-five years, Y
-
. Because it’s in your mind. This important tool is a<|endoftext|>Job Description: To Explore the Unknown⏎Curiosity about how life might have evolved on Mars could
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
- L5:436 suppresses
- L7:250 excites
- L8:235 suppresses
- L8:308 excites
- L8:937 excites
- L8:80 excites
- L7:362 suppresses
- L7:792 excites
- L8:2469 excites
- L8:731 weakly excites
- L6:150 weakly suppresses
- L8:494 weakly suppresses
- L8:157 weakly suppresses
- L7:260 weakly suppresses
- L7:4 weakly excites
- L4:366 weakly suppresses
- L8:322 weakly excites
- L8:2343 weakly suppresses
- L6:359 weakly excites
- L3:767 weakly suppresses
- L7:1420 weakly excites
- L7:301 weakly excites
- L6:63 weakly suppresses
- L7:506 weakly suppresses
Feeds later layers
- L10:936 suppresses
- L14:438 suppresses
- L10:1710 excites
- L15:3048 excites
- L11:1144 excites
- L13:1470 weakly suppresses
- L14:591 weakly excites
- L10:405 weakly suppresses
- L11:1197 weakly excites
- L14:2286 weakly suppresses
- L13:57 weakly suppresses
- L12:85 weakly suppresses
- L14:1268 weakly excites
- L14:2128 weakly suppresses
- L10:725 weakly suppresses
- L10:2676 weakly suppresses
- L21:1234 weakly excites
- L14:1743 weakly suppresses
- L12:1072 weakly suppresses
- L10:1132 weakly excites
- L10:2006 weakly suppresses
- L15:3051 weakly excites
- L10:2418 weakly suppresses
- L12:1338 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
- ми
- ptron
- deo
- สเต
- EK
- 하겠습니다
- #${
- comes
- theless
- #%
Suppresses
- opper
- olución
- atches
- aque
- сыл
- çi
- atement
- Symbols
- accomplish
- accompl