gap before a year in dates
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. 12 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 space right after a month and day, like 'born on August 1,', where the next thing is a year. When it was turned up, the model tended to write years such as 2005, 1929 and 2020.
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.
-
.⏎Yves Saint Laurent was a European fashion designer born on August 1, 1936 in Oran, Algeria. As a teen, he left
-
a charge of bad taste,<|endoftext|>8 edition of Principles of Biblical Interpretation found in the catalog.⏎April 1, 2003⏎by Baker Academic⏎Written in English⏎|The Physical
-
|⏎|Born||December 7, 1928⏎|Era||20th / 21
-
“thermostatic mechanism” that would counter-balance the dominant presence of electronic media in the lives of children.⏎How to Cite⏎Copyright (c) 2017 Robert Albrecht⏎Copyrights and publishing rights of all
-
province of Segovia, in 1559; died at Madrid, 27 March, 1625. He was a great-grandson of the Tordesillas
-
of the most beloved leaders India has ever produced.⏎Life history of Subhash Chandra Bose⏎Subhash Chandra Bose was born on 23 January, 1897 in Cuttack (Orissa) to Janakinath
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: and I have, but I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I have, I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
mynursingpaper.com/wp-content/uploads/2020/06/cropped-1559720319
-
rights to unionized or lack thereof).Please provide resources as required by APA standards.⏎https://mynursingpaper.com/wp-content/uploads/2020/06/cropped-1559720319
-
data. However, a means of observing the kinds of relationships between variables was needed. Using the theory of linear regression which originated from studies performed by Sir Francis Galton (1822-1911), the scatter diagram was developed
-
servant” are perhaps better understood as “slave,” in our times not an acceptable self-identity. Another difficulty: the actual lifestyles of servants/slaves in ancient cultures. Slaves were engaged by rich people to carry out their commands. Serv
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
- L11:2530 strongly excites
- L6:2811 suppresses
- L9:1460 excites
- L11:1975 suppresses
- L11:2229 excites
- L9:811 suppresses
- L9:1880 suppresses
- L8:457 excites year digits in copyright and dates
- L4:1291 excites second digit of years 20xx
- L4:1395 weakly suppresses
- L11:464 weakly suppresses
- L10:456 weakly suppresses
- L9:540 weakly suppresses
- L9:955 weakly excites
- L8:2335 weakly excites
- L9:534 weakly excites
- L9:1311 weakly suppresses
- L4:2321 weakly excites
- L9:382 weakly suppresses
- L8:2347 weakly suppresses
- L9:2452 weakly suppresses
- L11:1865 weakly excites open parenthesis in citations
- L1:2444 weakly suppresses
- L8:534 weakly suppresses
Feeds later layers
- L15:2236 excites
- L17:1476 excites
- L13:2285 excites
- L18:2103 excites
- L13:1342 excites digits inside years like 2017
- L15:1010 excites year digits inside 20xx dates
- L16:394 excites
- L18:1984 excites
- L14:1253 excites
- L17:1264 excites
- L18:2542 excites
- L14:1803 excites
- L19:1329 excites
- L14:1700 excites
- L18:1865 weakly excites
- L14:2374 weakly suppresses
- L19:2030 weakly excites
- L22:1279 weakly excites
- L14:809 weakly excites
- L25:2960 weakly excites
- L13:376 weakly excites
- L13:1322 weakly excites
- L15:622 weakly suppresses
- L14:2439 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
- 公元
- WXYZ
- 晚年
- 垓
- 近几年
- 公元前
- recent
- getObject
- 新的一年
- abis
Suppresses
- 指数
- .ct
- AGR
- 委员
- 篪
- rippling
- Indicator
- ăng
- cong
- hart