equals signs in maths and equations
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 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 equals sign in arithmetic and algebra, such as "24 = 4 times 6" or "15+19 =". Turning it up pushed the model toward number puzzles and counting, though the match is loose.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
, 24 = 4 ⋅ 6. So to solve the second part, on
-
- I'll do that 1/5 in purple -- this is going to be equal to 1/5 --⏎- I hope you don't hear that crow<|endoftext|>The
-
correct answer (ex: 15+19=??? or what is 12 divided by 3???). This game is
-
urgy celebrated?⏎- When is the liturgy celebrated?⏎- Where is the liturgy celebrated?<|endoftext|>A particle is moving along the parabola 4y = (x + 2)^2 in such a way that its x-coordinate is
-
,” the report said.<|endoftext|>Determining if a Product is Positive or Negative⏎Recall the following properties of multiplication of signed numbers:⏎(positive)(positive) = positive⏎(positive)(negative) = negative⏎(negative)(negative) =
-
– 39 = 611 candles.⏎Using a bar diagram⏎In the bar diagram,
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: it's not a problem that you can be a computer is not a problem that you can be a person is not a problem that you can be a robot is not a problem that you can be a person is not a person is not a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
François Mitterand who, in 1954, defiantly pronounced that ‘L’Algérie, c’est la France’. Nicolas Sarkozy, on the other hand, dramatically distanced himself
-
world’s most prosperous countries with strong international trading links (its port is one of the world’s busiest in terms of tonnage handled) and with per capita GDP equal
-
Important numbers!⏎xs = [6, 9, 13, 42]⏎-- Without library⏎f :: Bool -> [Int]⏎f b = if b then xs else ⏎-- With library⏎g :: Bool -> [Int
-
according to a report by the United Nations Population Fund.⏎Baby No 7 Billion will probably be born in the Asia-Pacific region, where the population growth rate is the highest in the world.⏎China's family planning policy, which limits most mainland couples
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
- L4:24 strongly excites
- L3:76 suppresses
- L0:423 suppresses
- L6:2237 excites
- L8:312 suppresses
- L4:2069 excites
- L8:469 suppresses
- L4:20 suppresses
- L4:1197 weakly suppresses
- L6:654 weakly excites
- L3:1213 weakly excites
- L7:1171 weakly excites
- L6:1037 weakly suppresses
- L6:1974 weakly excites
- L7:350 weakly excites
- L8:1650 weakly excites
- L1:1430 weakly suppresses
- L6:2065 weakly excites
- L6:777 weakly suppresses
- L8:137 weakly suppresses
- L5:1016 weakly excites
- L3:648 weakly suppresses
- L5:1030 weakly suppresses
- L6:564 weakly excites
Feeds later layers
- L17:1066 excites
- L10:2144 excites
- L10:647 excites
- L17:439 excites
- L18:841 excites
- L14:827 excites open bracket after a value or citation
- L11:2292 excites 'than' in comparisons
- L11:2066 excites
- L16:2028 excites
- L12:2171 excites
- L14:952 excites
- L13:287 suppresses
- L17:882 suppresses
- L15:1847 excites
- L11:681 excites
- L17:215 excites
- L12:2562 excites
- L10:205 suppresses
- L19:835 excites
- L10:206 excites colon or equals sign before a definition
- L14:924 excites
- L10:759 excites
- L14:1326 suppresses
- L12:2548 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
- immers
- 一定程度
- 的一员
- ensual
- るもの
- immer
- 某种程度
- 于是我
- oyer
- 协会会员
Suppresses
- .ts
- 他是
- ��
- ADM
- 岸
- atatype
- ('@
- ature
- ucc
- co