unclear: mixed, mostly technical prose
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
This neuron fires on a grab bag of positions in technical text, such as a blank before a digit, 'physically', 'costs' or 'energy'. No single theme stands out. Turning it up produced unrelated text.
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.
-
What is the centripetal acceleration of the same sphere if both distance R and speed s are doubled?⏎B) a / 4⏎C) a / 2⏎D) 2 a⏎E) 4 a⏎What is
-
obtain any kind of textual and visual information from the ever-growing online knowledge base through our computer or smart phone. For instance, when reading articles you get their corresponding text file executed, while when viewing images you have their image files displayed.
-
of products—agents that occlude dentin tubules and desensitizing agents that interfere with the transmission of nerve impulses.⏎Occluding agents act by physically covering or “plugging” the open, exposed dentin tubules, thus
-
small radiators in the functional spaces.⏎In a Passivhaus project, the value saved by avoiding a large central heating and distribution system, and its ongoing fuel costs, is invested instead in triple glazing (and sometimes, on north sides,
-
in your home, and watch the savings add up. Some people also prefer light emitting diodes (LEDs), which can be more expensive but use very little energy.⏎4. Energy-Conserving Building Envelope⏎Most houses leak air.
-
the VAT as well as the currency.⏎The task: Depending on the country and the data type you want to get the correct value from the table. For example, you select “Canada” and “VAT”, you want to get 5
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 following results in the case of a case of the case of the case of the case of the case of the case of the case of the case of the case of the case of the case of the case of the case of the case
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
measured.⏎Millimeter (abbreviated mm) – this is one thousandth of a meter. In scientific notation, 1 millimeter is 1 – E3 m, which means 0.001 x 1
-
A single heat pulse will switch the magnetic orientation of a nanometer-sized magnetic grain in just a few picoseconds (1 picosecond is 0.000000000
-
FLs). Installing one CFL saves $15 annually, according to the Rocky Mountain Institute, a nonprofit energy research group. Multiply that by the number of lamps and fixtures in your home, and watch the savings add up. Some people also prefer
-
has three parts, Part A, short answer, has 1 problem (12 points). Part B has 5 shorter problems (7 points each, so 35 points). Part C has 3 traditional problems (15 points each
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
- L13:2712 suppresses
- L13:868 suppresses
- L13:2805 excites
- L13:1339 suppresses
- L13:2470 suppresses
- L13:2140 excites
- L12:2817 excites
- L12:2942 suppresses
- L13:1667 excites
- L10:2193 excites
- L12:2436 excites
- L9:1555 suppresses
- L3:661 suppresses
- L13:1774 suppresses
- L13:1059 suppresses
- L13:2724 suppresses
- L12:1935 excites
- L12:2137 excites
- L13:864 suppresses
- L11:1894 weakly excites
- L13:2987 weakly suppresses
- L13:994 weakly excites
- L12:1375 weakly suppresses
- L13:2413 weakly excites
Feeds later layers
- L15:3026 suppresses
- L15:1481 suppresses
- L15:2487 suppresses
- L15:874 excites
- L18:1948 suppresses
- L16:858 excites
- L18:1446 excites
- L16:2337 suppresses
- L15:1445 suppresses
- L17:204 suppresses
- L15:922 suppresses
- L15:475 weakly excites
- L16:1265 weakly excites
- L19:2149 weakly suppresses
- L21:1216 weakly suppresses
- L15:427 weakly excites
- L16:422 weakly excites
- L18:435 weakly excites
- L16:2425 weakly excites
- L16:348 weakly excites
- L17:2179 weakly suppresses
- L15:2604 weakly excites
- L15:1346 weakly excites
- L15:2610 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
- [name
- 毋
- يد
- [source
- .Atoi
- berries
- خام
- ataires
- 白白
- 钮
Suppresses
- hä
- afka
- PRS
- lescope
- erview
- lingen
- ezier
- 动机
- 金字塔
- 最小