formal document and standards citations
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. 22 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 lights up on the spot where a formal reference, standard number or document term is about to appear, like a standard code, a legal citation or a report title. When it was turned down, the model kept talking about documents, and when it was turned up the text drifted to unrelated topics.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
,and I wrote this document.⏎QRcode is developed by DENSO Corporation. It is standardized by JIS and ISO.⏎This document is based on JIS X 0510(1999).⏎If you want to know
-
align with requirements such as HIPAA (Health Insurance Portability and Accountability Act) or GDPR (General Data Protection Regulation). This ensures
-
In addition, OSHA requires that SDS preparers provide specific minimum information as detailed in Appendix D of 29 CFR 1910.1200. The SDS prepar
-
law on June 15, 1976.⏎There are two main parts to what we call the HUD Code. The first part is called “Part 3280, Manufactured Home Construction and Safety Standards”. This concerns
-
18;15(2):57-61 DOI:http://dx.doi.org/10.4314/aas.v15i2.5⏎© 2018 Author. This work is
-
an SDS, the employer or designated person(s) should contact the manufacturer to obtain one.⏎Hazard Communication Standards⏎The Hazard Communication Standard (HCS) (29 CFR 1910.1200(g)), revised in
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: In the word "letter", the document is a word that is both a letter and a document. The word "letter" has 444444444444444444444
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Google Apps facilitates you to include a digital signature in header of the mails which are being sent from your domain, this will allow the receivers of
-
know you are in danger and call the police.⏎• Open your own savings account to increase your independence.⏎• Leave money, an extra set of keys, copies of important documents, and extra clothes with someone you trust.⏎DURING A VIO
-
aaight. As the dictionary indicates, aight’s colloquial English, so you should probably avoid dropping it into your college essays, newspaper columns, and legal documents. It looks to be the most recent evolution of the “all right”
-
who came to the United States before us, the people in our country who are still here, and the<|endoftext|>What Is a Warning Letter?⏎A warning letter is a letter addressed to an erring individual whether in a company of any industry, 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
- L10:1724 suppresses
- L6:2472 excites
- L10:1881 excites
- L8:1609 suppresses
- L11:626 excites
- L20:139 suppresses
- L20:1875 weakly suppresses
- L21:1654 weakly excites
- L13:1585 weakly excites
- L17:1646 weakly suppresses
- L21:522 weakly excites
- L20:1477 weakly suppresses
- L9:574 weakly excites
- L8:1274 weakly suppresses
- L10:1248 weakly excites
- L20:443 weakly excites
- L4:2152 weakly suppresses
- L21:2545 weakly suppresses
- L17:133 weakly suppresses
- L6:2789 weakly suppresses
- L7:2244 weakly suppresses
- L9:1785 weakly suppresses
- L6:1409 weakly suppresses
- L7:2480 weakly excites
Feeds later layers
- L27:1502 suppresses
- L24:575 suppresses
- L26:1107 suppresses
- L25:2989 suppresses
- L23:1372 suppresses
- L26:2985 suppresses
- L26:1878 suppresses
- L25:1028 suppresses
- L26:2658 suppresses
- L26:1350 suppresses
- L27:1109 suppresses
- L26:2876 suppresses
- L24:2330 suppresses
- L25:847 weakly suppresses
- L25:2969 weakly suppresses
- L23:1823 weakly suppresses
- L25:2765 weakly suppresses
- L27:2421 weakly suppresses
- L24:2980 weakly suppresses the word 'page' in references
- L27:1643 weakly suppresses
- L27:2984 weakly suppresses
- L27:560 weakly suppresses
- L23:2622 weakly suppresses
- L26:1811 weakly suppresses
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
- uhe
- berg
- onest
- 贞
- icolor
- rtrim
- رام
- urious
- URY
- ridge
Suppresses
- document
- 文档
- Document
- documents
- Document
- <Document
- document
- documents
- documento