"and" in subject-area lists
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 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 "and" inside short lists of topics or keywords, such as "Physics and Astronomy" in a list of subject areas or "Ceramics and Composites". Turning it up gave text that did not show this list pattern, so the steering adds little.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
|Journal||Classical and Quantum Gravity|⏎|State||Published - Jul 7 2003|⏎ASJC Scopus subject areas⏎- Physics and Astronomy (miscellaneous)<|endoftext|>Most computer users are familiar with the concept of spy
-
experimentally derived films. Good agreement between model and experiment was observed.⏎ASJC Scopus subject areas⏎- Electronic, Optical and Magnetic Materials⏎- Ceramics and Composites⏎- Condensed Matter Physics⏎- Materials Chemistry<|endoftext|>Humans are intimately
-
original location and eventually came to the Museum.⏎Social Tags [?]european art 1100-1500 [x] european decorative arts and sculpture [x] fransiscan monestary too [x] of
-
in a discontinuity analysis on observations near the borders between anglophone and francophone countries.⏎- Colonial rule⏎ASJC Scopus subject areas⏎- Economics and Econometrics⏎- Organizational Behavior and Human Resource Management<|endoftext|>- About ASTI
-
African American material and in particular the Civil Rights struggle.⏎In the History: UK section you can find out about the British Empire (Captives: Britain, Empire and The World 1600-1850, Linda Colley
-
he was appointed the head of KHAD, the secret police. Under Najibullah's control, it is claimed that KHAD arrested, tortured and executed tens of thousands of Afghans. Najibullah replaced Babrak
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 can't be in a place like a place where a person who who... something... something... I... something... something... I can't be... I can't be... I can't be... I can't be... something...
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
http://www.cal-ipc.org/ip/management/plant_profiles/Rumex_acetosella.php⏎Rumex acetosella (red sorrel, sheep sorrel)⏎Rumex acetosella (red sorrel, sheep
-
, physical destruction (curettage with eletrodesiccation) and Mohs Micrographic Surgery. Newer methods under investigation include photodynamic therapy and immunochemotherapy.⏎In 1935, Dr. Frederic
-
11). http://www.cdc<|endoftext|>|This article is about the assassin of pharaoh Cleopatra. For other uses, see Aya or Amun.|⏎Prior to establishing the Hidden Ones, Aya served as an
-
Similar to assets, liabilities can also be classified as current or long-term, depending on their repayment timeline.⏎Shareholders' equity, also referred to as net worth or owner's equity, represents the residual interest in the company's assets after deducting
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
- L12:2981 suppresses
- L12:1756 excites
- L10:74 excites
- L9:92 suppresses "and" or "or" joining two items
- L8:70 excites 'and' between stacked superlatives
- L4:57 suppresses
- L13:131 excites 'and' joining parallel phrases
- L10:1643 suppresses
- L12:1235 suppresses commas inside numbers and lists
- L13:1982 suppresses
- L11:406 suppresses
- L12:2071 excites
- L5:1469 suppresses
- L12:51 excites
- L13:371 suppresses
- L6:427 suppresses
- L13:977 suppresses
- L6:1630 excites
- L12:2057 suppresses
- L7:999 weakly suppresses
- L12:2409 weakly excites
- L9:738 weakly suppresses
- L12:1095 weakly suppresses
- L12:3006 weakly suppresses
Feeds later layers
- L16:1774 suppresses
- L15:631 suppresses
- L15:1340 suppresses
- L16:2733 suppresses
- L15:217 excites
- L17:84 excites
- L16:300 suppresses commas inside lists of nouns
- L16:2900 suppresses
- L15:2287 suppresses
- L15:1636 excites
- L17:2002 suppresses
- L15:2400 suppresses
- L21:735 suppresses 'or' in ranges and alternatives
- L15:2534 excites
- L15:1064 weakly suppresses
- L15:55 weakly suppresses
- L16:857 weakly excites
- L16:1346 weakly suppresses
- L17:3061 weakly suppresses
- L16:2209 weakly excites
- L16:717 weakly suppresses
- L16:1234 weakly suppresses
- L17:1234 weakly suppresses
- L15:124 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
- ựa
- ution
- olars
- 的学习
- S
- concentration
- 的责任
- ut
- 善
- ется
Suppresses
- azo
- agrams
- ombat
- robe
- obook
- convertible
- otron
- ву
- ditor
- theon