second digit of years 20xx
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. 4 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 '0' in the second half of a year written as 20 followed by a digit, like the '0' in 2010. The model did not show a clear theme when this neuron was turned up.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
analysis of algoritms⏎robert.kern at gmail.com⏎Fri Sep 10 00:10:22 CEST 2010⏎On 9/9/10 4:39
-
well as 551 Alpine lakes between <|endoftext|>Gonorrhoea - the drugs don't work⏎Published: 23rd Dec 2011 08:41:37⏎The prospect of unt
-
about this plant, please leave a comment. Comments have to be approved before they are shown here.<|endoftext|>Save water with wind power⏎21 March 2012⏎The power sector is the one of the world's biggest consumers of
-
I have some simple code example?⏎## Random fruit response # Lobster March 3 2012 import random verb = ['erratic', 'unbelievable', '
-
are at right angles<|endoftext|>The latest news from academia, regulators⏎research labs and other things of interest⏎Posted: September 29, 2006⏎Magnetic nanobridge brings two worlds together⏎(Nanower
-
problems we face in adulthood? (read more here)<|endoftext|>Задание от гостя:Размещено 13.09.2016 19:19
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: you have to be... But I have to be... But I have to be... But I have to be... But I have to be... But I have to be... But I have to be... But I have to be... But
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to the brain caused by an excess in salt intake.⏎Dr. Costantino Iadecola, of the Feil Family Brain and Mind Research Institute at Weill Cornell Medicine in New York, led the study which involved giving mice between eight and
-
schools reopening, relaxed pandemic precautions, and surging Delta variant infections.⏎Patients who receive their primary care internal medicine or pediatric care at Weill Cornell Medicine can also receive the vaccine, along with their flu vaccine at an upcoming
-
wellbeing of its constituents is also of interest, especially to some of the oldest theorists in this school, the Transcendentalists.⏎Thoreau, Whitman, and Emerson have served as inspiration for generations of non-conformists, such
-
one of the first American periodicals to recognize emerging American literature and culture, separate from European content.⏎The magazine also contained rare en<|endoftext|>F. A. Hayek, “Planning, Science and Freedom”, Nature, Nov. 15,
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
- L2:92 weakly excites
- L0:1529 weakly excites
- L3:512 weakly suppresses
- L2:75 weakly suppresses
- L2:752 weakly excites
- L2:3068 weakly suppresses
- L3:272 weakly excites
- L1:1960 weakly excites
- L3:1911 weakly excites
- L1:2497 weakly suppresses
- L3:456 weakly suppresses
- L3:282 weakly suppresses
- L2:111 weakly suppresses
- L0:1099 weakly suppresses
- L2:896 weakly suppresses
- L1:1460 weakly excites
- L2:503 weakly excites
- L3:50 weakly excites
- L2:1141 weakly excites
- L3:244 weakly excites
- L3:2954 weakly suppresses
- L1:1357 weakly excites
- L2:2283 weakly suppresses
- L2:1321 weakly suppresses
Feeds later layers
- L8:457 excites year digits in copyright and dates
- L18:1976 excites
- L12:1264 excites gap before a year in dates
- L9:929 weakly excites 'each' in 'each other'
- L7:1835 weakly suppresses
- L20:548 weakly excites
- L17:1050 weakly excites
- L6:2643 weakly excites
- L11:2368 weakly suppresses
- L7:1953 weakly suppresses
- L13:1536 weakly excites
- L7:676 weakly suppresses
- L14:197 weakly suppresses
- L6:215 weakly excites
- L10:1525 weakly excites
- L5:2502 weakly suppresses
- L5:572 weakly suppresses
- L5:10 weakly excites
- L20:3052 weakly suppresses
- L15:1940 weakly suppresses
- L23:1565 weakly suppresses
- L23:2100 weakly suppresses
- L26:1459 weakly excites
- L19:1273 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
- iaux
- oons
- igue
- omen
- ibur
- iture
- IOUS
- igans
- oreal
- ṇ
Suppresses
- .ConnectionString
- 技术服务
- maint
- 上课
- 对此
- prostitution
- strtol
- QT
- отношении
- categorical