opening bracket in math and definitions
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. 5 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 an opening bracket '(' that introduces a definition, a formula part or a list of alternatives, such as 'vertices (or things)' or 'roots (or zeros)'. Turning it up didn't produce a clear theme in the sample 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.
-
industry are able to sell,” 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⏎(n
-
"⏎- S: (v) displace, fire, give notice, can, dismiss, give the
-
1) and V = (- 1 , 2 , 1 ,- 1) are orthogonal vectors in R 4 . Write the vector X = (1 , 1 , 1 , 2) in the form X =
-
collaborative coding environment can be conveniently captured by a graph. The vertices (or things) are people, organizations, projects, and tickets. The edges (
-
point particle or string- particle or triangle- particle? No, Euler's formula has quantity (pi) which says me that the particle must be only a circle . Now I
-
version.View Full Document⏎Unformatted text preview: itself, and if ( ) ( ) lim lim x c x c h x L g x = = then ( ) lim x c f x exists and is equal to L. Three Special Limits
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: your family - The family In the context of Python programming, what is the purpose of this context? ** Let me try to understand this. I need to think about the Python programming language. Okay, let's see what I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
When, the glow of the reference matches with the hot object through an eye piece, that electric
-
()⏎- Automatic variables left on the stack of languages like C or garbage collected in languages like Java.⏎Byers, Simon. Scalable Exploitation of, and Responses to Information Leakage Through Hidden Data in Published Documents, AT&T Research, April
-
finding leaks that would otherwise be hard to detect by direct observation of methane concentrations in the air.⏎This project is developing remote sensing methods of detecting, discriminating, and mapping the effects of natural gas leaks from underground pipelines. The current focus is on
-
the lender will be assessing the borrower's ability to repay the loan based on a worst-case scenario interest rate. The calculation for individuals with irregular income may be different from those
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
- L3:276 excites
- L2:1684 weakly suppresses
- L3:182 weakly suppresses
- L4:542 weakly excites
- L3:1820 weakly suppresses
- L1:282 weakly excites
- L3:353 weakly suppresses
- L1:2078 weakly excites
- L4:1493 weakly suppresses
- L4:493 weakly suppresses
- L0:2626 weakly excites
- L4:458 weakly suppresses
- L1:1676 weakly excites
- L4:859 weakly excites
- L1:132 weakly excites
- L3:378 weakly excites
- L2:2124 weakly suppresses
- L4:2860 weakly suppresses
- L3:76 weakly suppresses
- L4:896 weakly suppresses
- L4:2145 weakly excites
- L0:440 weakly suppresses
- L2:559 weakly excites
- L3:37 weakly excites
Feeds later layers
- L12:1502 excites
- L6:144 excites
- L11:2191 excites unclear: brackets and parentheses in text
- L26:2195 weakly suppresses
- L6:1462 weakly excites
- L16:1967 weakly excites
- L24:848 weakly suppresses
- L7:795 weakly excites
- L15:2516 weakly excites
- L21:1628 weakly suppresses
- L9:796 weakly excites
- L6:2078 weakly suppresses
- L10:784 weakly excites
- L8:2184 weakly excites
- L14:3050 weakly suppresses
- L12:1287 weakly excites
- L11:1361 weakly excites
- L22:803 weakly suppresses
- L15:1517 weakly excites
- L6:323 weakly suppresses
- L26:1 weakly suppresses
- L25:947 weakly suppresses
- L7:2630 weakly suppresses
- L17:1004 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
- atsu
- ochond
- utan
- AffineTransform
- agen
- 也应该
- ზ
- 填补
- ually
- b
Suppresses
- quier
- éric
- 部分
- 깐
- igion
- ivity
- ificant
- heid
- egie
- 一侧