'known' after 'better' or 'also'
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. 19 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 word 'known' when it comes right after phrases like 'better known as' or 'also known as'. When it was turned up, the text drifted into repeating the word 'difference' over and over.
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.
-
ived people around the world. Arguably, the most famous and relevant of these compounds is 1,3,7-trimethylxanthine, better known as caffeine.⏎It’s pretty common knowledge that caffeine, at least in a very
-
, a tropical forest in Panama. Their method was simple: They drilled holes into the trees and waited.⏎Wsiclo and his<|endoftext|>Eric Weisz, better known to the world as Harry Houdini,
-
. Bruce Weinstein, better known to some as The Ethics Guy, listed five of them in his book Ethical
-
minated in the capture and destruction of Jerusalem. Titus became the Roman emperor in 79. He completed construction of the Flavian Amphitheatre, better known as the Colosseum, and opened it with ceremonies lasting more than
-
ilded Age architect who was a friend and admirer.⏎And it would be a surprise to almost everyone that the 94-by-94-foot laboratory, known as Wardenclyffe and envisioned by Tesla as a prototype for a stupend
-
, the sun god, wearing a crown of rays graces the obverse. This image was based upon the head of the great statue of Helios, better known as the Colossus of Rhodes, one of the Seven Wonders of the
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 difference between the difference and the difference. The difference is the difference from the difference from the difference, the difference from the difference from the difference, etc. The difference from the difference from the difference is the main difference. The difference from
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
What Are the Issues Concerning Empathy?⏎- What Is Compassion?⏎- What Are the Issues Concerning Compassion?⏎- Compassion vs Empathy: What Is the Difference?⏎What Is Empathy?⏎Empathy is the ability to put
-
!<|endoftext|>How to Graph a Cotangent Function⏎The parent graphs of tangent and cotangent are comparable because they both have asymptotes and x-intercepts. The only differences you can see are the values of theta where the asymptotes and x-inter
-
bellies and chest. They can have gray patches, however, which sometimes causes them to be mistaken for their cousin the gray fox. The best way to tell the difference is to check the tail: gray foxes have a black tipped tails<|endoftext|>
-
. To tell the truth slant is a way to get around the entire truth, without lying.⏎3. What is the major difference between the two expressions?⏎The differences between the two expressions are that one is very blatant and obvious while the other
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
- L17:2817 suppresses
- L11:363 excites
- L15:1952 suppresses
- L15:1585 suppresses
- L17:1834 suppresses
- L2:2246 excites
- L12:2285 suppresses
- L14:1773 weakly excites
- L16:3009 weakly suppresses
- L15:3011 weakly excites
- L15:1741 weakly excites
- L4:2520 weakly suppresses
- L6:1423 weakly suppresses
- L17:2458 weakly excites
- L18:1277 weakly suppresses
- L6:1608 weakly excites
- L7:2470 weakly excites
- L13:515 weakly suppresses
- L18:2654 weakly excites
- L14:2329 weakly suppresses
- L11:2664 weakly excites
- L17:1531 weakly suppresses
- L17:2773 weakly suppresses
- L9:96 weakly suppresses
Feeds later layers
- L27:2511 strongly suppresses
- L25:648 strongly suppresses
- L23:1955 suppresses
- L21:2185 suppresses
- L24:2557 suppresses
- L27:1903 suppresses
- L22:2143 suppresses
- L23:1897 suppresses
- L23:934 suppresses
- L21:2973 suppresses
- L23:1721 suppresses
- L22:1337 suppresses
- L22:1410 suppresses
- L21:2200 weakly suppresses
- L21:1955 weakly suppresses
- L20:1812 weakly suppresses
- L23:2589 weakly suppresses
- L22:1793 weakly suppresses
- L27:1881 weakly suppresses
- L21:1090 weakly suppresses
- L23:875 weakly suppresses
- L27:1492 weakly suppresses
- L20:795 weakly excites
- L21:2473 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
- UGIN
- iasm
- otos
- rette
- 他自己
- eled
- osos
- eper
- 打折
- (units
Suppresses
- difference
- differences
- Difference
- difference
- 差异
- Difference
- 区别
- Differences
- diferença
- 差别