question mark or line break after an FAQ heading
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. 12 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
Fires on the question mark or line break that ends a question heading, such as "What is embalming?", just before the answer starts. Turning it up gave text that was often repetitive, so the steering does not clearly show this role.
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.
-
second edition of the book is produced under the sponsorship of the Linguistic Society of America and the US National Museum of Language.<|endoftext|>What is Past Perfect Tense⏎The past perfect tense describes an action completed before a certain moment in the past as
-
. Their primary advantages were the use of Iron over bronze for weapons and tools and the use of Heavy Chariots in battle.⏎Are the Hittites Greek?⏎One of a group of Indo-European speaking peoples intrusive to Anatolia, the H
-
?⏎We looked online to see if there were any tours but couldn’t find any. Hope they offer them in the future. That hole<|endoftext|>What is Cloud Computing?⏎In simple terms, cloud computing is a technology used for storing and accessing data,
-
commonly referred to as a last name because it is usually placed at the end of a person’s full name, after any given names.⏎What exactly is a surname?⏎Also called last name . the name that a person has in common with other
-
, it is essential that Business Leaders know the AI and ML trends and can keep up with the pace and change of the business environment.⏎What are AI and ML?⏎Artificial Intelligence can be defined as a field of development of
-
', 'loganberry'] print "You are an" , random.choice (verb) , random.choice (fruit)⏎Any Tutorial Videos?<|endoftext|>What is globalisation?⏎Globalisation is the process by which the world is becoming increasingly interconnected as a result
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 number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
two different numbers cant have the same cube). Therefore, by Definition 1, is one-to-one....⏎View Full Document⏎- Spring '09<|endoftext|>part of speech: noun⏎definition 1
-
provide a proxy)⏎|Through your library|⏎References found in this work BETA⏎No references found.⏎Citations of this work BETA⏎No citations found.⏎Similar books and articles⏎Piero Antuono & Jan Beyer (1
-
, Guatemala, Florida, and France between 1789 and 1809.2 Much remains to be known about the St. [End Page 37] Dominguan diaspora in France and in the circum-Car
-
is to cover the fundamental concepts of computer graphics, and the concentration is on graphics in three dimensions.⏎Home page url⏎Download or read it online for free here:⏎by Fränk Zimmer - Wolke Verlagsges.⏎Pd (
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
- L10:259 excites
- L10:330 excites
- L10:84 suppresses
- L6:216 weakly excites
- L9:25 weakly excites
- L9:366 weakly excites
- L8:304 weakly suppresses
- L10:157 weakly suppresses end of document and web boilerplate
- L10:33 weakly suppresses
- L8:579 weakly excites
- L1:1813 weakly suppresses
- L8:2 weakly excites
- L5:21 weakly excites
- L7:884 weakly excites
- L11:775 weakly excites
- L8:88 weakly excites line breaks inside wrapped sentences
- L10:631 weakly excites
- L10:2985 weakly suppresses
- L6:0 weakly excites
- L4:125 weakly suppresses
- L9:568 weakly suppresses email address endings in examples
- L0:1617 weakly excites
- L9:1714 weakly excites
- L3:3 weakly suppresses
Feeds later layers
- L14:57 suppresses
- L13:2078 suppresses
- L14:304 suppresses
- L15:985 weakly excites line breaks inside wrapped text
- L14:1428 weakly excites
- L14:1321 weakly suppresses
- L14:3020 weakly suppresses
- L17:10 weakly suppresses
- L14:1082 weakly excites
- L13:2792 weakly suppresses
- L16:22 weakly suppresses
- L16:247 weakly suppresses
- L14:712 weakly excites
- L14:345 weakly suppresses
- L22:406 weakly suppresses
- L14:387 weakly excites
- L18:227 weakly suppresses
- L13:1572 weakly excites
- L15:227 weakly suppresses
- L15:615 weakly suppresses
- L15:550 weakly excites
- L15:1114 weakly suppresses
- L13:1343 weakly excites
- L14:166 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
- abouts
- THING
- 另有
- lining
- 型企业
- 連結
- ظه
- evice
- lined
Suppresses
- fühl
- pains
- för
- phận
- /gin
- \DependencyInjection
- .calls
- ordion
- esis
- forever