operators and symbols in code or formulas
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. 14 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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
It fires on symbols such as plus signs, slashes, question marks and single letters that sit inside code or maths formulas. It is a mixed bag of technical notation. Turning it down made the model get stuck in repeated brackets.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
*) memchr (str, 'p', strlen(str));⏎printf ("'p' found at position %d.\n", pch-str+1);⏎printf ("'p' not found.\n");⏎- Compare two blocks
-
unorm = sqrt(u(1)*u(1)+u(2)*u(2)); %calculate intermediate value⏎udot(1) = -u(1)/R + u(2) - un<|endoftext|>Thanks for visiting⏎
-
folder into which you want to install the free C++Builder 5 command line tool development system.⏎- From the bin directory of your installation: Add “c:\Borland\Bcc55\bin” to the existing path⏎
-
,⏎- times the derivative of sine of 5x with respect to 5x, that's just going to be cosine of 5x.⏎- If we want to write this in differential form, which is useful
-
find 386 pages of resources to learn more about Caddo history and culture, https://scholarworks.sfasu.edu/cgi/viewcontent.cgi?article=4012&context=ita⏎Visit the THC’s Learning
-
will need⏎- Calculator with statistical functions⏎Define the bel and decibel. The bel is defined mathematically as LB = log10 (P1/P0) where P1 and P0 are two quanties that are in 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 Last of the Last of the (The (The (The (The (The (The (The (The (The (The (The (The (The (The (The (The (The (The (The (The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
= talk_to_server(parse_xml_document=True)If you read it out loud, the code "reads"
-
1.5⏎UICBioE240 problem 1.18⏎Write a code that will follow the equation y = x * (x + x) * x.⏎Mile to Kilometer⏎Given A4 sizes find
-
= ['raspberry', 'pomegranite', 'bananana', 'grape', 'loganberry'] print "You are an" , random.choice (verb) , random.choice (fruit)⏎Any Tutorial Videos?<|endoftext|>What
-
in a safe manner, forceps need to be clamped on to the umbilical cord which has been cut. This cord, even after it is cut is connected to the woman’s
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
- L13:2254 suppresses
- L12:1576 excites
- L8:74 suppresses
- L5:2442 excites
- L12:473 excites
- L13:1855 suppresses opening parenthesis in citations and metadata
- L11:2697 weakly excites
- L11:2589 weakly suppresses
- L5:3043 weakly suppresses
- L9:2116 weakly suppresses
- L3:131 weakly suppresses
- L13:1395 weakly excites
- L12:2094 weakly suppresses
- L8:892 weakly suppresses
- L13:1691 weakly suppresses
- L13:1940 weakly excites
- L13:2088 weakly excites
- L7:5 weakly excites
- L3:2567 weakly excites
- L3:1348 weakly excites
- L10:239 weakly excites
- L2:2076 weakly suppresses
- L7:482 weakly excites
- L2:1780 weakly suppresses
Feeds later layers
- L15:1634 suppresses
- L17:783 suppresses
- L18:2023 suppresses
- L15:1857 suppresses
- L16:2306 suppresses
- L27:86 suppresses
- L26:22 suppresses the 'res' fragment inside words
- L18:1132 weakly suppresses
- L16:2949 weakly suppresses
- L18:671 weakly suppresses
- L19:1497 weakly suppresses
- L17:1959 weakly suppresses
- L16:601 weakly excites
- L17:488 weakly suppresses
- L16:1723 weakly suppresses
- L16:2503 weakly suppresses
- L18:115 weakly suppresses
- L16:1777 weakly suppresses
- L16:1254 weakly suppresses
- L15:2491 weakly suppresses
- L15:617 weakly suppresses
- L17:2861 weakly suppresses
- L15:1594 weakly suppresses
- L18:862 weakly excites hedges like 'it turns out' and 'if any'
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
- __:
- :date
- ended
- woff
- :A
- deserve
- .HOUR
- uchs
- :end
- ;,
Suppresses
- ()}
- {}{⏎
- concat
- ']}}</
- lient
- lian
- чь
- Guys
- 揮
- li