'see' cross-references and web boilerplate
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. 16 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 in website and reference text, mostly on words that point you somewhere else, like 'See' in 'See RAID 0', 'See more articles on', or 'click anywhere to close'. Turning it up made the model write in a more general, listy style, with no obvious link to the trigger.
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.
-
showing you is much cooler than explaining it.⏎Thanks for Subscribing!⏎You'll need to confirm your email, so please check your inbox.⏎(click anywhere to close)<|endoftext|>According to findings of a major HIV prevention trial presented Monday, March
-
© 2012, Columbia University Press. All rights reserved.⏎More on rheology from Fact Monster:⏎See more Encyclopedia articles on: Physics<|endoftext|>A startling statement is found in Mark 9:48. In
-
www.msfc.nasa.gov/news⏎The above story is based on materials provided by NASA's Marshall Space Flight Center. Note: Materials may be edited for content and length.⏎Cite This Page:<|endoftext|>By A. M. Sullivan⏎CHAPTER IX
-
consider the Requirements for RAID 5 Volumes and Guidelines for RAID 5 Volumes. Many striping guidelines also apply to RAID 5 volume configurations. See RAID 0 Volume Requirements.⏎A RAID 5 volume
-
the lower case letter L (for 'length'). For eaxmaple⏎would be read as "the length of the arc AB is 6 inches". See Arc Length page for more.⏎The angle measure is the angle formed by the arc at
-
to another dc voltage.⏎Current — See electric current.⏎DC — See<|endoftext|>Cherry Creek History⏎The Cherry Creek region is one of Nevada's oldest
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 like that, but you can't say that, or that; or that; or that; and so on; and so on; and so on. and so on; and so on; and so on; and so on.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
pneumonia. However, those living in the sunnier communities had lower death rates after developing influenza than those living in the less sunny communities. Vitamin D helped in two ways: first by reducing the aggressive immune response that among other things ruptured some of
-
bacteria that lead to pneumonia.⏎There is also good evidence that vitamin D reduces the risk of viral infections. Influenza is more common in winter for two reasons: cold temperatures and low solar UVB doses and vitamin D production. A randomized
-
the WHO Report on Global Tobacco Epidemic 2008 has recommended a six-point policy package called MPOWER for effective tobacco control. These policies are: (i) Monitor tobacco use and prevention policies (Article 20); (ii
-
This is likely due to a combination of factors, including Tesla’s higher battery capacity.⏎“There are several potential explanations for why EV
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
- L15:717 excites
- L14:667 suppresses
- L15:2871 suppresses
- L15:512 excites
- L15:3019 excites
- L15:427 suppresses
- L15:2610 excites
- L13:52 suppresses
- L15:1990 suppresses
- L15:917 suppresses
- L14:1333 suppresses
- L15:1119 excites
- L14:702 suppresses
- L15:746 excites
- L15:506 excites
- L15:2258 suppresses
- L10:2423 suppresses
- L15:551 suppresses
- L15:645 excites
- L13:19 suppresses
- L15:774 excites
- L13:3032 excites
- L12:207 suppresses
- L14:1939 suppresses
Feeds later layers
- L17:1088 strongly excites
- L18:1171 excites
- L17:933 excites
- L17:2648 excites number word before a list of reasons
- L17:2384 excites
- L17:1430 excites
- L18:1636 suppresses
- L20:631 excites
- L17:1718 excites
- L17:1267 excites
- L18:2151 excites
- L22:2193 excites
- L17:1930 excites
- L19:2057 suppresses
- L18:2837 excites
- L17:450 suppresses
- L18:1948 excites
- L18:1031 excites
- L18:2722 suppresses
- L19:2776 excites
- L22:1428 excites
- L21:1049 weakly excites
- L20:460 weakly excites
- L19:657 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
- ':⏎⏎
- fix
- ---⏎⏎
- 面前
- 什么呢
- exels
- przec
- 是什么呢
- ###
- ":⏎⏎
Suppresses
- <|endoftext|>
- otropic
- jure
- ouv
- ⏎ ⏎ ⏎
- acades
- vertising
- goog
- tré
- awns