the number '100' in numerals
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. 2 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 number 100, such as "100 years" or "10,000". Turning it up did not produce a consistent change in what the model wrote.
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.
-
the biggest collection of Educational Titanic resources and should be the starting point for any teacher and student. Their education section is very comprehensive.⏎National Geographic Titanic 100: A very current and fresh collection of Titanic content with a large amount of input
-
CCD reports.⏎Yet another factor is the ISO rating of your CCD. At ISO 400 you'll notice more hot pixels than at ISO 100, simply because the signal is amplified.⏎A hot pixel can range from bright white
-
backs, stock prices. One bizarre rumor about the AIDS virus was that it was developed in a Western laboratory and that the World Health Organization inoculated 100,000
-
along a 100-kilometer long north-south linear vent system over the past 5 million
-
function was checked again during four assessments over 11 years using the Modified Mini-Mental State examination (3MS), which is graded on a 100 point scale. A DASH diet adherence score was created based on consumption levels of
-
AbstractThe tobacco epidemic is one of the major global public health problems today that is being facilitated through a variety of complex factors. The tobacco epidemic killed 100 million people worldwide in the 20th century and it
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 lot of power, and I have a lot of power." So, I have a lot of power. What is the result of the first experiment, and what is the second thought? So, both sides of the first
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
such problems could prevent the potential development and perpetuation of psychological disorders among these individuals that may affect their efforts to address their obesity,” study researcher Marzieh Hosseini Araghi, BSc, MPH, of the University of Birmingham,
-
instance, "hok" warns of serious aerial threats -- usually eagles -- whereas "hok-
-
of 15,795 razorback sucker and 11,696 bonytail were released with 134.2 kilohertz (kHz)
-
ives, in areas of skin away from the bite.⏎• Hoarseness or shortness of breath.⏎Anaphylaxis requires emergency medical attention. People
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
- L0:697 excites the last zero in round numbers
- L0:920 weakly excites digit '9' inside years and numbers
- L1:288 weakly excites
- L1:2330 weakly excites
- L1:178 weakly suppresses
- L1:2310 weakly excites
- L0:1741 weakly suppresses
- L1:1900 weakly excites
- L1:544 weakly excites
- L0:1541 weakly excites
- L0:369 weakly excites
- L1:565 weakly excites
- L0:529 weakly suppresses
- L0:1121 weakly excites
- L1:3022 weakly suppresses
- L1:293 weakly suppresses
- L1:2772 weakly excites
- L1:2884 weakly excites
- L1:2414 weakly excites
- L0:85 weakly suppresses
- L1:50 weakly excites
- L1:2385 weakly excites
- L0:2594 weakly excites
- L0:2108 weakly suppresses
Feeds later layers
- L23:1099 suppresses
- L20:117 weakly excites
- L8:2995 weakly suppresses
- L23:869 weakly excites
- L8:3067 weakly suppresses
- L7:2031 weakly suppresses
- L7:2098 weakly excites
- L7:2541 weakly suppresses
- L22:1509 weakly excites
- L18:2231 weakly suppresses
- L4:505 weakly suppresses
- L20:145 weakly excites
- L5:73 weakly suppresses
- L10:424 weakly suppresses
- L17:410 weakly excites
- L19:1851 weakly excites
- L10:813 weakly excites
- L26:859 weakly excites
- L9:307 weakly suppresses
- L23:2138 weakly suppresses
- L25:1310 weakly excites
- L6:715 weakly excites
- L14:2954 weakly excites
- L12:1909 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
- izzie
- 让用户
- śni
- <tag
- orama
- .Companion
- ousse
- urance
- ushi
- [tag
Suppresses
- �
- press
- -offs
- reserv
- pair
- ACT
- significant
- �
- relocation
- olumes