"in" after "shown" (figure references)
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. 21 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 responds to the word "in" when it follows "shown" or "as shown" and points to a figure, as in "as shown in the figure". Turning it up made the model repeat phrases like "as shown in the figure" over and over.
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.
-
the linear movement as far as possible. Based on the analysis of the tooth surface cutting movement, the revolution and rotation of the blade are further analyzed, as shown in the figure.<|endoftext|>Source: The Independent – independent.co.uk⏎Up, Up
-
sample. It is a very sensitive technique that requires a very high vacuum (10-9~10-11 mbar). As schematically shown in Figure 1, a pulsed beam of primary ions (usually Cs or Ga
-
several phases on the stator. The rotor is toothed and made of metal, but is not permanently magnetized. A simplified variable reluctance stepping motor is shown in Figure 2. In this figure, the rotor has four teeth and the st
-
as Poisson’s equation. Geometry is assumed to be circular and windings are modeled as current sheets placed on the core surface. The induction machine geometry is shown in Figure 2. Poisson’s equation is solved analytically for geometry and
-
.⏎One of the most common applications of lead screws is for positioning the print head in 3D printers. They often use 8mm lead screws as demonstrated in the video below. The concepts explained above are all shown in the first few seconds
-
colored glass (usually red), to make lights monochromatic. The lamp is connected to a battery source through an ammeter and a rheostat as shown in the figure.⏎The optical pyrometer works in a certain simple
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 a 300-100-100-100-100 - 100 - 100 - 100 - 100 - 100
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
10, p. 252) considered them older fluvio-glacial drifts. Thomson (1917, p. 411) refers to them more fully, but is extremely
-
ability; (2) stability over time; (3) affect on individual behaviour, and; (4) sense of shared values between the members (Ibid.).⏎Old institutionalism focus on the formal institutions, meaning law and government,3
-
for viewers moving through the decorated house, arguing that memory played a vital role in the creation and reception of Roman pictorial ensembles in domestic situations.”Ibid., 225. As I explore the continued vitality of the household memory theater
-
firepower over and over again.” William Rosenau, Book Note, The Origins of the First Modern Weapon, TECH. REV., Jan. 1987, at 74, (reviewing John Ellis,
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
- L16:1709 excites
- L20:3063 suppresses
- L20:440 suppresses
- L19:1490 excites
- L19:943 excites
- L18:1101 excites
- L17:2311 excites
- L20:388 suppresses
- L14:1411 excites
- L3:860 suppresses
- L17:2176 suppresses
- L17:1072 weakly suppresses
- L14:227 weakly excites
- L20:2462 weakly excites
- L19:2654 weakly excites
- L18:2656 weakly excites
- L20:223 weakly excites
- L20:1679 weakly suppresses
- L18:2047 weakly suppresses
- L8:933 weakly suppresses words introducing an example or aside
- L20:1413 weakly excites
- L8:1796 weakly suppresses
- L19:579 weakly suppresses
- L3:286 weakly excites
Feeds later layers
- L27:2066 strongly excites
- L26:2455 strongly excites
- L23:1387 excites
- L22:1385 excites
- L23:886 excites
- L24:2043 excites
- L27:2275 excites
- L26:2509 excites
- L22:1593 excites
- L22:1794 excites "shown in" before a figure reference
- L23:1453 excites
- L22:960 suppresses
- L27:1122 excites
- L24:1154 excites
- L22:1900 excites
- L27:2283 excites
- L23:374 weakly excites
- L26:1020 weakly excites
- L27:1913 weakly excites
- L27:1276 weakly excites
- L25:2290 weakly excites
- L24:2854 weakly suppresses
- L25:791 weakly excites
- L24:1704 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
- Figure
- Figure
- figure
- figure
- Fig
- Fig
- Figures
- fig
- 图
- FIG
Suppresses
- quán
- ộng
- wald
- EmailAddress
- groupName
- Taste
- zá
- izable
- democratic
- 漱