colon after 'Published' in bylines
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. 8 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 fires on the colon right after the word 'Published' in article headers, such as 'Published: September 2020'. When it is turned up, the text does not clearly change in a way that matches this.
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.
-
four million, which would imply that reaching the fourth million occurred in less than thirty years.⏎Autor: gf Ana L. Dávila Román⏎Published: September 20,
-
could be challenged.⏎"British culture was extremely strait-laced in the 1950s<|endoftext|>By ANAHAD O'CONNOR⏎Published: November 21, 2006⏎THE CLAIM -- Pricking
-
Science Grants by Hughes Institute⏎By WILLIAM CELIS 3d⏎Published: October 4, 1994⏎An $86 million grant
-
famous Krimml waterfalls; as well as 551 Alpine lakes between <|endoftext|>Gonorrhoea - the drugs don't work⏎Published: 23rd Dec 2011 08:41
-
that policies mandating free water access in schools may help to improve excellence in access, and more work is needed to help all schools excel in this area.⏎Published: September 2020⏎ID #: 73248⏎
-
0.3390/ijerph9114170⏎Published: 15 November 2012⏎Abstract: Understanding the effects of
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 a new blog at: ... and so on: ... So, for example, I have a new blog at: ... The problem is, I have a new one: So, I have a new one: The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
GNU Oflox⏎This is a GNU package:octave⏎released on 31 May 2012⏎|License||Verified by||Verified on||Notes|⏎|GPLv2||Kelly Hopkins||6 December
-
1939 – 4 June 1948⏎|Preceded by||James Barry Munnik Hertzog|⏎|Succeeded by||Daniel François Malan|⏎3 September 1919 – 3
-
across the US, this is worth a read. The full article can be accessed here.⏎Tags: New York Times, Insect Population Decline<|endoftext|>First posted on Myth & Moor in 2015.⏎"In Britain there were once hundreds
-
.-H. (2014, July). Effectiveness of a tiered intervention model in Chinese: A growth model of reading fluency. Paper presented at the Twenty-First Annual Meeting: Society for the Scientific Study of Reading, Santa
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
- L7:52 excites
- L7:1784 weakly excites punctuation right after 'of' or 'with'
- L6:12 weakly excites
- L7:547 weakly suppresses
- L0:3005 weakly excites
- L4:568 weakly excites
- L4:3026 weakly suppresses
- L5:2005 weakly excites
- L7:2074 weakly excites
- L7:162 weakly suppresses
- L5:94 weakly suppresses
- L7:3011 weakly excites
- L5:96 weakly suppresses
- L0:1023 weakly excites
- L7:2204 weakly suppresses
- L6:1136 weakly excites
- L6:18 weakly excites
- L5:384 weakly excites
- L4:378 weakly suppresses
- L7:1266 weakly suppresses
- L3:514 weakly suppresses
- L1:2683 weakly excites
- L3:187 weakly excites
- L7:191 weakly excites
Feeds later layers
- L9:2417 suppresses
- L9:1724 suppresses
- L9:326 weakly suppresses
- L10:125 weakly suppresses
- L11:58 weakly excites web address colon in urls
- L23:2183 weakly excites
- L14:1292 weakly suppresses
- L22:2255 weakly suppresses
- L16:2833 weakly suppresses
- L13:2993 weakly suppresses
- L10:2570 weakly suppresses
- L10:105 weakly excites
- L27:1097 weakly excites
- L25:1304 weakly excites
- L19:2409 weakly suppresses
- L23:676 weakly excites the 'm' in a.m. and p.m
- L22:2907 weakly suppresses
- L14:710 weakly suppresses
- L9:82 weakly suppresses
- L19:2327 weakly suppresses
- L10:2583 weakly suppresses
- L9:270 weakly suppresses
- L26:340 weakly suppresses
- L22:361 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
- gemäß
- //}⏎⏎
- //}⏎
- mute
- icers
- .Misc
- nuts
- вали
- niest
- comparable
Suppresses
- :@
- ensor
- suming
- 關係
- otion
- 濡
- à
- عنه
- :
- 传感器