placeholder example email domain
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. 1 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 '@example' part of made-up email addresses like 'firstname.lastname@example.org'. Its strongest negative reactions are on 'art' in phrases like 'state of the art', which is a different pattern. Turning it up did not produce text about email or contact details.
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 Department of Construction, Chaoyang University of Technology, Taichung 41349, Taiwan (e-mail: firstname.lastname@example.org).⏎Cite:Ming-Ju Lee, Ming-Gin Lee, Y
-
of Construction, Chaoyang University of Technology, Taichung 41349, Taiwan (corresponding author's e-mail: firstname.lastname@example.org, email@example.com).⏎Chia-Liang Chiang is with
-
version be performed in public only once; if interested in presenting a realization of the A.N.L. - folds, please contact Sever Tipei (firstname.lastname@example.org) to obtain a personal version.⏎Five variants of the ANL-folds
-
: Czech Centre for Science and Society⏎Contact person: RNDr. Karel Charvát, firstname.lastname@example.org⏎Visit Inspire4Youth’s NEW Facebook page!⏎Open INSPI
-
the publication for benchmarking.⏎To explore opportunities for collaboration with subject-matter experts on your assessment and evaluation efforts, contact PRA Well-Being at firstname.lastname@example.org.⏎This resource was first shared in 2019.<|endoftext|>Social
-
how your school can get involved with Bush Builders contact firstname.lastname@example.org⏎Brothers Island tuatara⏎For just over six years, Wellington
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: the main idea. This is a possible way to think about the problem. This is a possible way to think about the problem. I will now present the key results. This is a possible way to think about the problem. This is a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
determine the three dimensional structure and function of biomolecules at atomic resolution. This is significantly important for researching the basic components of life and understanding illnesses. State of the art experimental stations for serial femtosecond crystallography offer the opportunity to
-
0 days – nearly twice the duration achieved by state-of-the-art NASA balloons. “Even three weeks is very rare,” Lindstrand scoffed.⏎
-
for electric vehicles is still a work in progress however. The first genuine electric motor was produced as early as 1834. Today, state-of-the-art, innovative technologies are still based on discoveries made by researchers nearly<|endoftext|>The material in
-
was made primarily for use in calculators . By today's standards, this microprocessor would widely be considered inadequate, but at the time it was state-of-the-art. The 4-bit processor was made up of 2,<|endoftext|>The European
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
- L0:1224 suppresses
- L0:178 suppresses
- L0:1301 suppresses
- L0:893 weakly excites
- L0:34 weakly excites
- L0:1150 weakly suppresses
- L0:281 weakly excites
- L0:166 weakly suppresses
- L0:2074 weakly suppresses
- L0:420 weakly excites
- L0:2962 weakly excites
- L0:1807 weakly suppresses
- L0:1633 weakly excites
- L0:262 weakly excites
- L0:1063 weakly suppresses
- L0:1404 weakly excites
- L0:1626 weakly suppresses
- L0:378 weakly excites
- L0:2745 weakly excites
- L0:1818 weakly suppresses
- L0:1266 weakly excites
- L0:691 weakly suppresses
- L0:1153 weakly excites
- L0:1190 weakly excites
Feeds later layers
- L14:2170 weakly suppresses
- L22:990 weakly suppresses
- L11:1729 weakly suppresses
- L12:466 weakly excites
- L15:544 weakly excites
- L5:2826 weakly suppresses
- L8:327 weakly suppresses
- L3:200 weakly suppresses comma after a finished clause
- L16:1057 weakly excites
- L24:581 weakly suppresses
- L14:1890 weakly excites
- L23:1411 weakly excites
- L7:1392 weakly suppresses
- L18:2119 weakly suppresses
- L2:1032 weakly excites
- L13:2523 weakly suppresses
- L23:1081 weakly suppresses
- L27:61 weakly excites
- L22:664 weakly suppresses
- L18:1286 weakly suppresses
- L2:1040 weakly suppresses
- L21:2289 weakly suppresses
- L9:1013 weakly excites
- L8:1568 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
- 党的领导
- 度过
- idente
- pains
- 筋
- 스트
- .ByteString
- 主义思想
- _MetadataUsageId
- 济
Suppresses
- itionally
- bard
- ories
- ourd
- ллер
- 婵
- ulu
- rema
- вел
- ceu