'al' in 'et al.' citations
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. 19 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 'al' in academic citations like 'Krebs et al., 1995'. It also reacts weakly to odd symbols in web addresses. Steering changed the topic a lot without any clear link to citations.
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.
-
regulation of prey numbers, population control of mesopredators through competition, and maintenance of a functional balance of biodiversity in the local community (Krebs et al., 1995; Terborgh et al., 19
-
08) is associated with marked ecological changes, including alterations to climatic regimes, phenology, resource availability and habitat quality (Zhou et al. 2004; Grimm et al. 2008
-
Floral humidity, an area of elevated humidity within the headspace of the flower, has been demonstrated to occur in a number of flower species (Corbet et al., 1979
-
European beech (Fagus sylvatica L.) are dominant tree species, accounting for 30% of the forest areas (Pretzsch et al. 2014). To elucidate the drought responses of these two
-
habitat, human-caused mortality, and isolation of small populations are major concerns in the conservation of large carnivores (Clark et al., 1996; Singleton & Lehmkuhl, 2
-
Hymenoptera: Vespidae) are widely distributed around most of the globe and are diverse and common in many landscapes (Beggs et al., 2011). Paper wasps are likely to influence many other
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 to be able to take the project in the present, but to do that, I have to know the future is coming, but in the future, the person is expected to have the knowledge to know the future, which is a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
archive.jpr.org.uk/10.1177%2F13591045050537
-
and%20adaptive%20immune%20system.⏎- How Vitamin C Supports a Healthy Imm
-
> <H1>This is the Page Heading</H1> This is the first paragraph. <P> This is another paragraph
-
Python treats escape sequences in strings as special commands. For example, the `t` character inside a string prints a tab character (equivalent to four spaces):⏎s = “NametAgetMarks”⏎Name<|endoftext|>FAQ About Fashion⏎Cout
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
- L18:1836 suppresses
- L15:1451 excites
- L9:722 suppresses
- L9:502 suppresses
- L4:2315 weakly suppresses
- L18:542 weakly suppresses
- L13:1411 weakly excites
- L12:2837 weakly excites
- L11:927 weakly excites
- L18:1128 weakly excites
- L0:148 weakly excites
- L12:95 weakly suppresses
- L15:2628 weakly excites
- L15:500 weakly excites
- L4:2208 weakly excites
- L11:1213 weakly excites
- L9:2228 weakly suppresses
- L8:2705 weakly excites
- L18:350 weakly excites
- L16:679 weakly suppresses
- L9:35 weakly excites
- L3:1278 weakly excites
- L2:573 weakly suppresses
- L2:2772 weakly suppresses
Feeds later layers
- L20:1092 excites
- L21:137 excites
- L20:573 excites
- L22:709 weakly excites
- L21:2775 weakly excites
- L20:2293 weakly excites
- L27:2426 weakly excites
- L21:322 weakly excites
- L27:452 weakly excites
- L22:1311 weakly suppresses
- L23:1945 weakly excites
- L22:856 weakly excites
- L23:2286 weakly excites
- L25:1398 weakly excites
- L27:2936 weakly excites
- L24:1850 weakly excites
- L26:2914 weakly excites
- L26:534 weakly excites
- L24:1697 weakly excites
- L22:2075 weakly excites
- L20:3034 weakly excites
- L20:2160 weakly excites
- L22:2383 weakly excites
- L20:2093 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
- rowser
- IZED
- ized
- icrobial
- iddle
-
- ISED
- IDDLE
- ruled
- ум
Suppresses
- DataContext
- /bind
- ystate
- Paw
- Dane
- trys
- 实践
- técnica
- 表白
- _va