signs and constants in 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. 10 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
This neuron fires on plus and minus signs and small numbers inside maths formulas, such as the '+ 1' in n + 1 or the '2' in a parabola equation. It is weak and mixed, so it is hard to say more than that. Turning it up did not produce a clear change in the replies.
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.
-
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 increasing at a constant rate
-
5 Y 2 . Solution : X =- 5 W c) Find some solution of A X = 3 Y
-
⏎Atomic Symbol: Rn⏎Atomic Weight: 222⏎Melting Point: -95 F (-71 C)⏎Boiling Point: -79 F (-61.7 C)⏎Word origin: The word rad
-
are parallel.⏎Hence, the given lines are parallel.⏎Let's consider another set of equations.⏎y = 2x +3 and 4y = –2x+10⏎On comparing y = 2x +3 and
-
` initialized to 1, increment⏎it each round through the loop, and check it before running the loop. When the⏎check indicates that it equals n + 1 (
-
) you will obtain your answer:⏎F'(x) = 2x√(1+x6)⏎Comment. There is no need to integrate and then differentiate by 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 have been using the following code: (A, B) where A and B are the values, and I used the values, and the code that I can be used for a small number of (A, B) is the first
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
- Filed Under⏎J ennifer Skaja has seen the signs of small miracles, as children
-
CO2 + H2O.⏎How does baking soda and vinegar make a chemical reaction?⏎Can you
-
generally share the communitarian position on issues relating to the economy, such as the need for environmental protection and public education, but not on cultural<|endoftext|>Cite This Page⏎Louisa Musgrove Timeline and Summary⏎- Louisa and her
-
to explore STEM topics in a way that deepens their knowledge while also developing 21st-century skills.⏎Read the rest of this entry »⏎Filed under: Gaming, Services, Software, 3 million dollar grant, Associate Professor, Aug
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
- L9:1988 strongly suppresses
- L7:678 suppresses
- L6:1187 suppresses
- L8:1305 weakly suppresses
- L9:327 weakly excites
- L4:722 weakly excites
- L9:589 weakly excites
- L4:1056 weakly excites
- L0:1133 weakly suppresses
- L8:1162 weakly suppresses
- L3:2196 weakly suppresses
- L9:1249 weakly excites
- L7:2569 weakly suppresses
- L4:2927 weakly excites
- L9:1834 weakly excites
- L5:163 weakly suppresses
- L9:2414 weakly suppresses
- L8:1217 weakly excites
- L4:1947 weakly suppresses
- L4:2622 weakly excites
- L1:1582 weakly excites
- L3:2127 weakly excites
- L3:2093 weakly excites
- L5:1475 weakly excites
Feeds later layers
- L18:137 suppresses 'the' after 'to' in fixed phrases
- L17:399 suppresses
- L11:689 excites
- L20:164 suppresses
- L18:365 suppresses
- L11:1515 excites
- L16:1066 suppresses
- L17:765 weakly suppresses
- L18:344 weakly excites
- L11:1659 weakly excites
- L19:643 weakly suppresses
- L27:672 weakly suppresses
- L22:2254 weakly suppresses
- L18:404 weakly excites 'get', 'go' and 'got' after a subject
- L26:2674 weakly suppresses
- L27:2672 weakly suppresses word-start 't' fragment in split words
- L26:2002 weakly excites the "un" prefix splitting a word
- L26:2552 weakly suppresses
- L22:316 weakly suppresses
- L16:827 weakly suppresses
- L24:1837 weakly excites
- L20:1070 weakly excites
- L25:189 weakly suppresses
- L27:1887 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
- aption
- waves
- oods
- forcer
- went
- ueva
- rica
- ackson
- 口水
- wife
Suppresses
- té
- ied
- UnderTest
- antage
- 첨
- /jav
- contemplate
- ertiary
- juan
- typeid