'information' after words like 'provides' or 'for'
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. 17 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 word "information" when it follows phrases like "provides", "looking for" or "all the". When it was turned up, the text drifted into vague, general talk, with no clear link to the word.
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.
-
from the Windows 8 Store directly to your computer.⏎Check out this article from our archives to learn more about apps: What Are Apps?<|endoftext|>This guide provides information how to apply the ecosystem approach to a project or issue. It does not set
-
but the party will be hoping the world sees images of orderly troop formations and dancing<|endoftext|>Carpal Tunnel Home > Carpel Tunnel Syndrome⏎Were you looking for information about Carpal Tunnel Syndrome? Carpel tunnel syndrome is a common misspelling
-
with actual tips and suggestions here’s some information to SSL certificates from their early days.⏎It was March 1995
-
a writing folder as a<|endoftext|>This page provides information about the questions in Sundanese, also called the interrogative form. This
-
giving you all the information you wanted about the art. Welcome to augmented reality. Virtual reality replaces the real
-
CA.⏎Other Resources on Rumex acetosella⏎- USDA PLANTS database - Federal database with information on identification and distribution, and links to websites in individual states.⏎- Jep
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 have been a bit of a common place name of a word of words of words of words of words of words of words of of of of of of of of of of of of of of of of of of of of of of of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
-Flint)⏎What is a CMS and how can it be used?⏎A CMS (content management system) is a piece of software used to effectively organize and manage large amounts of digital content. Content, in reference to the
-
content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Neither BioOne nor the owners and publishers of the content make, and they explicitly disclaim, any express or implied representations or
-
.<|endoftext|>Start a 10-Day Free Trial to Unlock the Full Review⏎Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections.
-
Cloud Computing?⏎In simple terms, cloud computing is a technology used for storing and accessing data, application and programs over the Internet. Cloud computing refers to the delivery of hosted and computing services including servers, storage, databases, networking, analytics, and
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
- L8:1403 weakly suppresses
- L7:1359 weakly excites
- L12:1692 weakly excites
- L11:2296 weakly excites
- L14:1684 weakly excites
- L16:200 weakly excites
- L15:2814 weakly suppresses
- L10:609 weakly excites
- L8:1106 weakly excites
- L13:10 weakly suppresses
- L16:1716 weakly excites
- L5:1700 weakly suppresses
- L0:3044 weakly suppresses
- L5:229 weakly excites
- L13:2255 weakly suppresses
- L11:1972 weakly suppresses
- L14:924 weakly suppresses
- L9:2340 weakly excites
- L11:2964 weakly suppresses
- L5:2282 weakly suppresses
- L2:1147 weakly suppresses
- L16:222 weakly suppresses
- L7:2722 weakly suppresses
- L16:715 weakly excites
Feeds later layers
- L20:643 excites
- L19:1517 suppresses
- L18:2651 excites
- L18:2530 weakly excites
- L22:524 weakly excites
- L18:2422 weakly excites
- L24:2034 weakly excites
- L24:2877 weakly suppresses
- L18:285 weakly suppresses
- L18:1277 weakly suppresses
- L26:1708 weakly excites
- L20:2005 weakly excites
- L24:1203 weakly excites
- L20:2407 weakly excites
- L21:1999 weakly suppresses 'well' in compound adjectives
- L27:3071 weakly suppresses
- L25:2818 weakly suppresses
- L18:919 weakly excites
- L19:504 weakly excites
- L21:2890 weakly excites
- L18:1567 weakly suppresses
- L27:2008 weakly suppresses
- L23:1159 weakly suppresses
- L18:699 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
- ��
- 匮乏
- centers
- สำค
- ocalypse
- _BS
- 被捕
- หมด
- bout
- ién
Suppresses
- cial
- imes
- 械
- eger
- ward
- 替
- зам
- ıyor
- juana
- OLUTE