space between month and day
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. 6 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 blank space between a month name and the day number, as in 'April 22' or 'July 20 1969'. Its opposite reaction is on the last digit of a year, like the '9' in '1969'. Turning it up did not produce a clear theme, so steering doesn't confirm the pattern. Its activations are fairly small compared with the other neurons here.
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.
-
similar study model<|endoftext|>Celebrate Earth Day 2012Certification Map, Education, General Interest | No Comments »⏎Earth Day is celebrated on April 22, and this year marks the 42nd anniversary
-
1863. In the two World Wars, Fort Hamilton served as a major embarkation and separation center. Most recently, in the aftermath of the September 11th terrorist attacks, National Guard troops stationed at Fort Hamilton provided critical operational
-
a Research Paper | Pen and the Pad There are two types of research<|endoftext|>The Apollo 11 space flight landed the first humans on Earth’s Moon on July 20 1969. The mission, carried out by the United
-
a man who would go on to become a leading member of the National Left Wing Council and editor of that group's Chicago newspaper, The Communist.⏎Following the September 1918 state convention of the Socialist Party of Michigan, a number of
-
“weekends” are not recognised.⏎|Sunday||February 05||Kashmir Day||Observed since 1990
-
view at the Portland Museum of Art January 24, 2008, through March 23, 2008.⏎The January 20, 2008, edition of The Boston Globe features an
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 would like to know what I would like to know for the future. I want to know what I want to know for the next year. I want to know what I want to know for the next year. So, I want to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
1969 to 51 percent in the most recent estimates.⏎|Sources of Income -
-
also argue the Internet (<|endoftext|>このトピックには0件の返信が含まれ、1人の参加者がいます。2020年7月6日 at 9:45 AMに Megan Smith さんが最後の更新
-
1978)⏎- National Medal of Science (1986); the APA’s Award
-
1928 noted that for children taking vitamin D, there were fewer cavities and the cavity
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
- L4:342 suppresses
- L2:24 suppresses
- L5:7 weakly suppresses
- L4:76 weakly excites
- L3:187 weakly excites
- L3:2576 weakly suppresses
- L4:1171 weakly excites
- L1:699 weakly excites
- L3:130 weakly excites
- L5:129 weakly excites
- L0:1587 weakly suppresses
- L5:703 weakly excites
- L4:2850 weakly excites
- L1:1855 weakly suppresses
- L0:856 weakly excites
- L5:1923 weakly excites
- L5:1939 weakly excites
- L2:493 weakly suppresses
- L1:1203 weakly excites
- L4:330 weakly suppresses
- L4:1441 weakly suppresses
- L4:2896 weakly excites
- L0:2118 weakly excites
- L3:637 weakly excites
Feeds later layers
- L13:1155 strongly suppresses
- L7:101 strongly suppresses
- L16:715 strongly suppresses
- L8:423 suppresses
- L11:173 suppresses
- L14:248 suppresses
- L17:163 suppresses
- L18:1321 suppresses
- L13:74 suppresses
- L21:498 suppresses
- L13:2099 suppresses
- L16:1655 suppresses
- L16:662 excites
- L20:1197 suppresses 'this' pointing at the current document
- L9:1430 suppresses
- L12:1065 suppresses
- L21:278 suppresses
- L21:708 suppresses
- L21:68 excites
- L12:1755 excites
- L20:272 suppresses
- L17:1121 suppresses
- L17:390 suppresses
- L15:668 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
- жить
- idel
- pez
- <Resource
- 莎
- フル
- affles
- unde
- [res
- mina
Suppresses
- 型
- ية
- 性
- 度
- 型企业
- 性和
- 者
- type
- 关系
- s