conservation laws and invariance
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. 4 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 responds to physics wording about things that stay the same, such as "conservation", "invariant" and "conserved". Turning it up pushed the text toward maths and physics ideas, though not always coherently.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
. An invariant profile was observed when movements were normalized for velocity and distance. Two arms, extended
-
Noether's theorem, which states that there is a one-to-one correspondence between each one of them and a differentiable symmetry of nature. For example, the conservation of energy follows from the time-invariance of physical systems, and the conservation
-
to symmetries and invariance principles, three special conservation laws have been described, associated with inversion or reversal of space, time, and charge.⏎Conservation laws are considered to be fundamental laws of nature, with broad application in physics,
-
, which apply to such quantities as mass, parity, lepton number, baryon number, strangeness, hypercharge, etc. These quantities are conserved in certain classes of physics processes, but not in all.⏎A local conservation law
-
negative numbers⏎<|endoftext|>In physics, a conservation law states that a particular measurable property of an isolated physical system does not change as the system evolves over time. Exact conservation laws include conservation of energy, conservation of linear momentum, conservation of angular momentum,
-
change form. In general, the total quantity of the property governed by that law remains unchanged during physical processes. With respect to classical physics, conservation laws include conservation of energy, mass (or matter), linear momentum, angular momentum, and electric charge
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: (A) A (B) A (C) A (D) A Let me think about this. The question is about a certain item. Which of the following is the correct one? A. A
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
involvement.⏎Introduction. Part 1. Geometry and Nature. Chaos game visualization of sequences (H. J. Jeffrey). Tumor growth simulation (W
-
equal.⏎Label this plum tree graph to make it totally magic!⏎A graph is a set of vertices (or nodes), together with a set of edges (or arcs).⏎Look at
-
⏎From symmetric meson theory, the formation of helium nucleus from two deuterons, i.e., fusion, requires necessarily a direct force due to exchange of
-
the⏎major arc AB,⏎where you go the long way around the bottom of the circle. Unless stated otherwise, it always means the⏎minor arc - the shortest of the two.⏎If you want to indicate the major arc
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
- L3:105 excites
- L3:25 suppresses
- L3:798 suppresses
- L3:50 excites
- L0:178 suppresses
- L0:1224 suppresses
- L3:52 weakly suppresses
- L3:91 weakly excites
- L0:1301 weakly suppresses
- L3:448 weakly suppresses
- L3:1990 weakly suppresses
- L2:1899 weakly excites
- L1:1464 weakly excites
- L3:556 weakly suppresses
- L3:421 weakly suppresses
- L3:2850 weakly suppresses
- L3:115 weakly suppresses
- L2:1809 weakly excites
- L2:2349 weakly suppresses word pieces inside cinnamon, China, cabar
- L3:102 weakly suppresses
- L3:185 weakly suppresses
- L2:1334 weakly suppresses
- L2:160 weakly suppresses
- L2:2759 weakly excites
Feeds later layers
- L7:2422 excites
- L10:1001 weakly excites
- L12:786 weakly excites
- L8:1934 weakly excites
- L17:2082 weakly suppresses
- L9:477 weakly excites
- L7:968 weakly excites
- L5:2894 weakly excites
- L9:417 weakly excites
- L11:2354 weakly excites
- L13:1246 weakly suppresses
- L9:1673 weakly excites
- L11:2307 weakly excites
- L16:1732 weakly excites
- L12:1259 weakly suppresses
- L8:912 weakly excites
- L16:2051 weakly suppresses
- L21:659 weakly suppresses end of document boundary
- L9:1959 weakly excites
- L23:875 weakly suppresses
- L5:1881 weakly excites
- L11:2468 weakly excites
- L10:2774 weakly excites
- L17:1005 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
- OLLOW
- nants
- 功率
- давать
- inciple
- 틴
- iterate
- 纲
- transit
- →super
Suppresses
- istical
- quirer
- mock
- opard
- zzle
- isnan
- OMET
- ostringstream
- azio
- izzo