opening parenthesis in citations and metadata
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. 13 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 lights up on the opening bracket in citation style text, like the bracket after 'Copyright' or after a journal name. Turning it up made the model drift into odd, rambling text with no clear theme.
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.
-
hoped that this paper is a kickstart that will give a useful input to those interested in Artificial Intelligence in education.⏎|Subjects:||L Education > L Education (General)|⏎|Deposited By:||Zaidatun Tasir
-
Valentine, Harry T.;⏎Source: Journal Agricultural, Biological, and Environmental Statistics 10:460-477⏎Publication Series: Scientific Journal (JRNL)⏎Description: Transects consisting of multiple, connected segments with a prescribed
-
captive crocodilians Authors⏎|Davis, Lisa - APPLIED BIOSYSTEMS INC|⏎Submitted to: Journal of Experimental Zoology⏎Publication Type: Peer Reviewed Journal⏎Publication Acceptance Date: February 1
-
water, were the major factors shaping phytoplankton communities.⏎- Abstract views: 1777⏎- PDF: 842⏎Copyright (c) 2009 Silvia TAVERNINI, Daniele N
-
, student academic ability, and students learning in their mother tongue are some factors for consideration. Copyright © 2014 Society for the Scientific Study of Reading (SSSR).⏎|Publication status||Published - Jul
-
of creating a “thermostatic mechanism” that would counter-balance the dominant presence of electronic media in the lives of children.⏎How to Cite⏎Copyright (c) 2017 Robert Albrecht⏎Copyrights and publishing
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 problem with the problem that I have to do this in the school that I am doing my job, and I need to help the students in the school, and I am responsible for the students and the school in my job in my work
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
function multiplied by its conjugate. The statement of the evenness of the porobability
-
table with a specific name or index exists.⏎The following example retrieves a single table by name from the DataTableCollection.⏎Private Sub GetTableByName() ' Presuming a DataGrid is displaying more than one table
-
conuts and banana leaves<|endoftext|>const void * memchr ( const void * ptr, int value, size_t num );⏎
-
code, and humans infer things from the code that will be necessary to work on it.⏎typedef void * model_3d_ref;⏎model_3d_ref load_model_from_disk(const char * absolute_file_path);⏎void draw_model(model_3
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
- L11:746 excites
- L7:158 excites
- L12:1196 suppresses
- L11:1024 weakly suppresses
- L11:1912 weakly excites words right after an 'if' condition
- L12:2762 weakly suppresses
- L11:2581 weakly excites
- L11:1927 weakly excites
- L11:3048 weakly excites
- L5:897 weakly suppresses
- L11:540 weakly suppresses
- L10:1157 weakly suppresses
- L3:102 weakly suppresses
- L3:488 weakly excites
- L3:173 weakly suppresses
- L3:1607 weakly excites
- L10:296 weakly excites
- L6:1036 weakly suppresses
- L9:1302 weakly suppresses
- L8:344 weakly suppresses
- L7:656 weakly suppresses
- L12:120 weakly excites
- L11:1479 weakly excites
- L9:1366 weakly suppresses
Feeds later layers
- L18:1501 suppresses
- L16:2306 suppresses
- L16:1777 suppresses
- L15:1857 suppresses
- L15:1634 suppresses
- L19:369 suppresses
- L14:2958 suppresses operators and symbols in code or formulas
- L26:1331 weakly suppresses
- L25:360 weakly suppresses
- L20:758 weakly suppresses
- L16:2482 weakly excites
- L14:1559 weakly suppresses
- L23:224 weakly suppresses
- L18:1266 weakly suppresses
- L15:2582 weakly excites
- L19:900 weakly suppresses opening parenthesis after a term
- L16:590 weakly excites
- L26:476 weakly excites
- L16:2116 weakly suppresses
- L16:508 weakly suppresses
- L20:2956 weakly suppresses
- L14:696 weakly suppresses
- L18:2141 weakly excites
- L15:2335 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
- =yes
- dozen
- --; ⏎
- --;⏎
- .ZERO
- .LENGTH
- ches
- DBus
- /Math
- ��
Suppresses
- lage
- 而在
- 圭
- ícia
- ervention
- điển
- лез
- bare
- rząd
- prática