end of document token after a finished passage
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
This neuron mostly lights up at the marker that separates one document from the next, especially after a passage has finished its last sentence. It also fires on a few cut-off words at the end of a text. Turning it up made the writing repetitive and dull.
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.
-
was packed with (hidden) learning.⏎Here's how the completed assignment looks:⏎Six little math problems, scissors and glue, and a whole lot of learning.<|endoftext|>Popular Inventions
-
to have no cost. An example of free is a place where citizens may speak and act as they please, as long as they respect this right of others.<|endoftext|>Bifunctionals show that sometimes simply tethering two useful molecules together unlocks some
-
Exposure Attention Interpretation Exposure Exposure highlights the stimulus to which a consumer is exposed to. Exposure can be either deliberate or random. ...⏎Not exactly what you need?<|endoftext|>Allegany County’s
-
uous clothing; equipment; outfit.⏎- To cover with a caparison, as a horse.⏎- To dress sumptuously; adorn with rich dress.<|endoftext|>The idea of transgenerational epigenetic inheritance of acquired behaviors is in the
-
least 12 missions to carry cargo to and from the International Space Station as part of the Commercial Resupply Services (CRS) contract for NASA.⏎For<|endoftext|>|1st person||me me||nos us|⏎|2nd person||
-
the use of technologies in the process of training.⏎the task like that was defined as a task of a multicriteria choice under conditions of uncertainty.⏎To compare the<|endoftext|>A Friday story in National Geographic reports that NASA has found a way to restart the
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 meaning of the word s I have to put into the kind of things I have to put into the kind of things I had to put into the kind of things I had to put into the kind of things. So I need to put
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. "Sir G. C. Whitlock (During the 1834 Coorg Campaign)". www.archerfamily.org.uk. Archer Family. Archived from the original on 13 November 2013. Retrieved 1
-
can tolerate both full sun and partial shade.<|endoftext|>Copyright © University of Cambridge. All rights reserved.⏎'The Solid' printed from http://nrich.maths.org/⏎is one of a series of problems designed to develop learners' team⏎working
-
uggle for Power in the Ussr. StudyMode.com. Retrieved 05, 2012, from http://www.studymode.com/essays/The-Struggle-For-Power-In-The-99
-
instruction in recognizing alternative interpretations in their writing. (A sample story and four figures of data are included.) (RS)⏎Publication Type: Speeches/Meeting Papers; Reports - Research⏎Education Level: N/A⏎Authoring Institution: N/A
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
- L10:58 excites
- L8:1060 suppresses
- L10:1144 suppresses sentence openers after a break
- L10:196 suppresses
- L0:2172 weakly suppresses
- L10:553 weakly suppresses list-style essay phrasing, weak
- L5:323 weakly excites
- L11:29 weakly excites
- L4:21 weakly excites
- L4:1564 weakly suppresses
- L3:1596 weakly suppresses
- L10:518 weakly excites
- L9:248 weakly excites
- L8:2010 weakly suppresses
- L10:351 weakly excites
- L9:948 weakly suppresses
- L9:1573 weakly suppresses
- L4:2052 weakly excites
- L5:1770 weakly suppresses
- L0:754 weakly suppresses
- L0:2363 weakly suppresses
- L3:2218 weakly suppresses
- L3:272 weakly excites
- L4:623 weakly suppresses
Feeds later layers
- L14:2288 suppresses
- L18:2100 excites
- L13:1001 excites
- L13:2328 weakly excites
- L13:1513 weakly suppresses
- L15:1299 weakly excites
- L19:802 weakly excites unclear: word before a company or brand name
- L14:586 weakly excites
- L13:2766 weakly excites
- L25:351 weakly suppresses
- L25:239 weakly suppresses
- L15:1936 weakly excites
- L14:2556 weakly suppresses
- L17:892 weakly suppresses
- L15:2628 weakly suppresses
- L18:1075 weakly suppresses
- L18:1136 weakly suppresses
- L23:1196 weakly suppresses
- L27:1984 weakly excites
- L19:2054 weakly suppresses
- L21:121 weakly excites
- L17:516 weakly excites
- L26:320 weakly excites
- L21:1856 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
- eteria
- eson
- lesia
- esy
- roring
- ament
- esan
- OffsetTable
- erral
- kea
Suppresses
- 一流的
- 不断的
- nice
- “⏎⏎
- ~
- led
- nuclei
- mond
- !”
- ~~