second number in a pair
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. 10 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 spot right before the second half of a pair of numbers, like the '95' in 'by 90 percent and 95 percent' or the '48' in 'of 36 or 48 volts'. It is tracking a number that comes next in a set of two. Turning it up made the text wander, so the steering results do not show a clear link.
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.
-
and HENCE v.i=10. LOCAL VARIABLE i is assigned value 0 on entry to method another() so printing these values gives 10 and 0 respectively. 3)The original object ref v is unchanged , it
-
the past 20 years, and numbers of antelope species, such as hartebeest and oryx, have declined by 90 percent and 95 percent
-
Designed as the third project in a series of Point-in-Time Surveys (PITS) (the previous two occurred in 1993-4 and 2001-2), this project surveyed Great Lakes residents over the age
-
type⏎cd example.com and it will look for⏎/var/www/example.com, stopping whenever it finds a match⏎or runs out of search locations.⏎You can have as many base search paths as you
-
climate change, the electric power industry’s emissions of carbon dioxide, CO2, have risen 5.9 percent since 2002 and 11.7 percent since 1997, the analysis shows.⏎
-
Koothu or two pairs of Koothus such as Vassai Koothu-Pukal Koothu; Vethiyal Koothu-Pothuviyal Koothu; Vari Koothu - Vari Santh
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 first time, I have been given a number and the first number is 2 for the first time. So I have been given a number and the first number is 2. So I have been given a number and the first number
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16 25.........⏎Basics: Divide integers
-
) and V = (- 1 , 2 , 1 ,- 1) are orthogonal vectors in R 4 . Write the vector X = (1 , 1 , 1 , 2) in the form X = a U
-
ants, six corporals, two buglers, and forty-two privates. A squadron was to be formed in each of the towns of Guerrero, Mier, Camargo, and Reinosa; and a company was to be organized at L
-
Following is the sequence:⏎- Chinta (Anxiety)⏎- Asha (Hope)⏎- Shraddha (Reverential belief, Faith, Virtue of being
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
- L9:587 excites
- L9:622 suppresses
- L8:719 excites
- L8:150 excites
- L8:780 excites
- L7:761 weakly suppresses
- L5:540 weakly excites
- L9:1517 weakly suppresses
- L5:1523 weakly excites
- L7:737 weakly suppresses comma after a place name
- L8:607 weakly excites
- L6:775 weakly suppresses
- L2:2173 weakly suppresses
- L9:1224 weakly excites
- L3:1996 weakly excites
- L7:519 weakly suppresses
- L6:835 weakly excites
- L9:1992 weakly excites
- L8:47 weakly excites
- L7:2772 weakly suppresses
- L0:98 weakly suppresses
- L4:1139 weakly excites
- L0:57 weakly excites
- L5:640 weakly suppresses
Feeds later layers
- L11:1777 excites
- L14:988 excites the word 'two' after 'the other'
- L14:1629 suppresses
- L11:1551 excites
- L13:1081 suppresses
- L14:2455 weakly excites
- L12:1814 weakly suppresses
- L24:2829 weakly excites
- L18:851 weakly suppresses
- L15:2545 weakly suppresses
- L12:690 weakly suppresses
- L12:2981 weakly excites
- L23:2509 weakly excites
- L13:650 weakly excites
- L15:1097 weakly excites
- L11:1090 weakly suppresses
- L11:2842 weakly excites
- L21:703 weakly excites
- L21:1742 weakly excites
- L24:849 weakly suppresses
- L15:1492 weakly suppresses
- L26:2627 weakly excites
- L15:1067 weakly suppresses
- L25:2692 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
- MBED
- irectional
- łam
- mouseleave
- �
- ustin
- IDL
- cestor
- EDIUM
- !");⏎
Suppresses
- @
- lar
- hostel
- requis
- 您好
- uja
- ActivityResult
- order
- repeats
- ObjectContext