'by' before a byline or agent
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. 9 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 'by' when it introduces a writer's name, like 'Posted ... by Michael Marsh', or when it introduces who did something. Steering it up made the text repeat itself, so the link to the effect is weak.
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.
-
code from Marcus Mock exam 2: following code what will be the output?⏎I added UBB code here - Mapraputa [This message has been edited by Mapraputa Is (edited
-
.3 light years.⏎The seventh-closest is a star most denizens of Earth would recognise: Sirius (alpha Canis Majoris), with eighth place taken by its companion Sirius B, sometimes referred to as 'the Pup'. Sirius B
-
ser is established, and will form again the following year. ®<|endoftext|>Posted: Sep 28, 2009 8:19 PM by Michael Marsh⏎Updated: Sep 28, 2009
-
tech gadgets that will protect your home against air pollutants<|endoftext|>Active and Passive Voice – Tips for Content Writing⏎February 22nd, 2012 by Donna Mathews⏎The use of active voice is important while writing online content.
-
11:37 am by Matt Prescott<|endoftext|>A significant fraction (~ 30%) of the gamma-ray sources
-
affect the performance of the candle in any way and also means your wax<|endoftext|>Equalization - Sound Mixing.⏎May 25, 2004 - by Franck ERNOULDTranslated by Dominique Zbiegiel.⏎In this
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 don't know where I am, but I know that I am not just a tourist. I know that I am not a tourist. I think the most important thing about the time I am being asked to provide information is to the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
great joy that will be for all the people. 11 For unto you is born this day in the city of David a Savior, who is Christ the Lord.
-
municipalities that fall within the district of the court (a geographical area that is included in the district court´s catchment area, usually a number of municipalities).⏎Who do you come into contact with at the district courts?⏎The officers serving at the district
-
why it is so significant that the angels described Jesus’ birth as “peace on earth.”⏎9 And an angel of the Lord appeared to them, and the glory of the Lord shone around them, and they were filled with great fear.
-
tcp connection is open. Protocol numbers internet assigned numbers authority.⏎Protocol cisco commands ios configuration 61 cheat sheets tagged with network. This number is assigned automatically by the os, manually by the. They are associated with tcpip network connections 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
- L6:1637 strongly excites
- L7:194 excites
- L6:2756 excites
- L8:1494 suppresses preposition right after a passive verb
- L6:2937 excites 'by' after a title or author credit
- L4:1394 excites
- L6:742 excites
- L6:1690 excites
- L6:408 suppresses
- L8:1041 suppresses
- L6:3045 excites
- L8:351 excites 'in' after 'high' or 'dressed'
- L7:360 suppresses
- L7:2766 suppresses
- L5:2745 suppresses
- L6:3070 weakly excites
- L8:400 weakly excites
- L6:1114 weakly excites
- L4:2646 weakly excites
- L6:1997 weakly suppresses
- L1:2414 weakly suppresses
- L5:2890 weakly suppresses
- L8:2656 weakly suppresses
- L6:716 weakly excites
Feeds later layers
- L15:2005 excites
- L18:56 excites
- L13:2832 excites "by" after a passive verb
- L12:1286 excites
- L12:2887 excites
- L12:755 excites
- L13:2465 excites
- L12:2535 excites
- L13:1291 excites
- L17:1221 excites
- L10:481 excites word before a noun naming people or a role
- L11:3015 excites
- L12:2266 excites
- L14:2374 excites
- L14:1796 excites
- L25:343 excites
- L25:1291 excites
- L11:853 excites
- L23:436 excites
- L16:2208 excites
- L19:2885 excites
- L10:362 excites
- L22:712 suppresses
- L17:505 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
- 从业者
- 队长
- etter
- emand
- Holder
- someone
- 艺术家
- 群体
- ainers
- professionals
Suppresses
- ��
- (xi
- (nx
- 主持召开
- Middle
- (ti
- track
- .TryParse
- ailles
- aim