the number 1 as a unit value
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 number 1 when it means a standard or normalized amount, like a speed of 1, a cosine of 1, or a value set to unity. It seems to pick up on 'one unit' meanings more than on the digit itself. Turning it up did not produce any clear theme in the written text.
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.
-
2pi). a) Using its own inner impulse (h) circle - particle moves ( as a wheel) in a straight line with constant speed c = 1. We call such particle - ?photon?.
-
constraint and characterize the solution to her intertemporal maximization problem.⏎c) Consider the full competitive market equilibrium in this economy; normalize the constant labor force to unity such that kt=Kt at all t. Characterize the equilibrium dynamics of
-
x = L/ 6?⏎Answer to relevant QuestionsWhich has the lowest (ground-state) energy, an electron trapped in a one dimensional box of length 10-6 m or one with length 10-10 m?
-
an HTML image tag. It displays a 1×1 transparent image then proceeds to execute PHP scripts based on the pseudocron schedule file
-
a given color that differs slightly from a primary color; "after several trials he mixed the shade of pink that she wanted"⏎t
-
(gas or liquid) from a zone of lower external concentration of same species in the groundwater. In mathematical terms, the net amount of material diffusing across a unit cross-section (J, flux) perpendicular to a membrane of known thickness (x
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 don't think that...". "The following is a list of people...": "This is a list of people..." "This is a list of people": "Message:" "Messages": "This is a list of people
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
services and evaluate video quality in six categories of digital defects on a scale of one to five (one being unwatchable and five being spectacular).”⏎“In addition
-
It typically features a touch plate known as "Sonic Strings", preset rhythms, auto-bass line functionality, and buttons for major, minor, and 7th chords. The most basic method of playing the instrument is to press the chord buttons
-
acid or something like that, the association will be greatly weakened or eliminated.⏎Some biological structures (for example, "promoters") are not as sensitive to base pair sequences as proteins are to amino acid sequences, but these are exceptional; we are
-
use this pattern so often that we have created a name and a symbol for it. 8×7×6×5×4×3×2×1 is written as 8! and we call it "eight factorial." You probably
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
- L8:184 suppresses
- L9:1373 suppresses
- L6:902 excites
- L9:1348 weakly suppresses
- L4:1100 weakly excites
- L5:513 weakly suppresses
- L7:1524 weakly suppresses
- L8:879 weakly excites
- L3:244 weakly suppresses
- L7:1499 weakly excites
- L9:1989 weakly excites
- L9:1691 weakly suppresses
- L9:2915 weakly suppresses
- L7:2156 weakly excites
- L5:303 weakly excites
- L9:1169 weakly suppresses
- L8:2742 weakly suppresses
- L8:1701 weakly suppresses
- L4:2090 weakly excites
- L7:1986 weakly excites
- L9:673 weakly suppresses
- L0:969 weakly excites
- L8:318 weakly excites
- L7:1947 weakly excites
Feeds later layers
- L11:871 suppresses zero and infinity as values
- L11:1852 excites the digit 1 in numbers
- L11:2291 excites numbers inside citation references
- L11:836 excites
- L17:1615 weakly suppresses
- L13:1479 weakly excites
- L18:2732 weakly suppresses
- L23:2219 weakly suppresses
- L12:2434 weakly excites
- L16:1012 weakly excites 'one' before 'of' in 'is one of the'
- L17:1154 weakly excites
- L20:547 weakly suppresses
- L16:1564 weakly suppresses
- L12:303 weakly suppresses
- L11:1509 weakly excites
- L16:2624 weakly excites
- L26:1882 weakly excites
- L20:2976 weakly suppresses
- L16:536 weakly suppresses
- L22:1639 weakly suppresses
- L13:1106 weakly excites
- L11:694 weakly suppresses
- L12:894 weakly excites
- L25:2345 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
- lox
- ştur
- Apprent
- Magical
- =BitConverter
- sağlam
- wij
- _AHB
- itarian
- urre
Suppresses
- holes
- handles
- asl
- segments
- OM
- segment
- essel
- SE
- ses
- rs