empty slot before a figure number
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. 3 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 fires on the spot right after the word "Figure" or "Fig." in science writing, where the figure number is expected next. Turning it up did not produce any clear theme in the model's writing.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. 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.
-
cellular processes was cemented with the awarding of the 2004 Nobel Prize in Chemistry to ubiquitin research (3).Ubiquitin (Figure 1) can be attached to substrate proteins as a single moiety or in the
-
had proposed various methods of crowning that can be achieved in the process of gear generation. Maag engineers have used crowning for making longitudinal corrections (Fig. 1a); modifying involute tooth
-
is in place for dairy factory employees and persons considered at risk after outbreaks in ruminants.⏎During 2005–2007 (Figure 1), a total of 105 human cases of brucellos
-
RNAs were extracted from specimens using TRIzol reagent. Qiagen's miScript System was used to synthesize cDNA and perform qPCR (Figure 1). Endogenous miRNAs are not polyadenylated, therefore during
-
2 km. The infrasound data is collected at 20 hz, while the meteorologic data is recorded at 1 hz.⏎Figure 4. Array layout. Blue dots indicate the 7 elements, while the purple
-
ech (c. 90 years old) and spruce (c. 70 years old) trees at the opposing sides of the plot (Figure 1a, for details see Grams et al. 2021
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: it is not, but I have to be the one, and the one is the one. It is not, but I have to be the one, and the one is the one. It is not, but I have to be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Resources to set limits that preserves the fishery. Indian fishing rights issues in the Northwest have been widely publicized over the years.⏎Alongside fishing rights, Amerindians also have the right to set up casinos and, much to my surprise,
-
0 to 1,000 years.⏎“What we can do is look at light bands in tree rings and correlate that to snowpack,” said Matheus. “Looking at snow, rather than runoff, is an important distinction because it
-
restricted by the installation of Road Speed Limiters. Road Speed Limiters were introduced in Europe in commercial trucks and passenger service vehicles initially to curb the maximum speed as this was seen as a major threat to road safety. While enforcing speed limits through fines
-
108], and in Arabic – in form ﺭﺎﻬ<|endoftext|>How Do I Score Well on the Spanish Praxis™?⏎The Spanish Praxis™ exam, more accurately known as the Praxis II®: Spanish: World
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
- L0:2224 weakly suppresses
- L2:2328 weakly suppresses
- L2:513 weakly suppresses
- L2:1 weakly suppresses
- L0:786 weakly suppresses
- L0:1529 weakly excites
- L0:2189 weakly excites
- L2:1999 weakly excites
- L2:59 weakly excites
- L0:771 weakly excites
- L2:1927 weakly excites
- L0:3063 weakly excites
- L1:1373 weakly suppresses
- L0:1259 weakly excites
- L2:2500 weakly suppresses
- L1:904 weakly suppresses
- L0:2815 weakly suppresses
- L1:1779 weakly suppresses
- L0:2731 weakly excites
- L0:1102 weakly excites
- L1:323 weakly excites
- L2:36 weakly excites
- L2:2382 weakly excites
- L1:255 weakly excites "to" after "the best way is"
Feeds later layers
- L8:1186 suppresses
- L4:2839 suppresses
- L6:20 weakly suppresses
- L4:2950 weakly suppresses
- L17:1053 weakly suppresses
- L17:1438 weakly suppresses
- L9:450 weakly suppresses
- L21:2701 weakly suppresses
- L4:1802 weakly suppresses
- L11:176 weakly excites
- L6:1260 weakly suppresses
- L7:2638 weakly suppresses
- L13:665 weakly excites
- L23:2547 weakly suppresses
- L4:761 weakly suppresses
- L23:1634 weakly suppresses
- L24:1169 weakly suppresses
- L4:2043 weakly suppresses
- L16:761 weakly excites superlatives like 'largest' and 'first'
- L8:2853 weakly suppresses
- L4:604 weakly suppresses
- L8:1549 weakly suppresses
- L12:2056 weakly suppresses
- L15:150 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
- ographed
- iculo
- icture
- .colorbar
- aghan
- ricula
- _sha
- quam
- ric
- illary
Suppresses
- scores
- 러
- NAS
- IEWS
- rat
- シーズン
- (HTTP
- ",⏎⏎
- '],['
- öst