email addresses at the end of text
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. 3 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 spot inside an email address, right before the ".org" part, in example addresses like firstname.lastname@example.org. When it is turned up, the model's writing drifts into unrelated topics like lists, numbered quizzes and news stories.
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.
-
engineering disciplines in India. He has also authored multiple books on various areas of data science. He may be reached at email@example.com (preferred) or firstname.lastname@example.org .<|endoftext|>Recovery from Addiction⏎There can be many different pathways to rem
-
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
-
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
-
wide variety of languages, catering for beginning students to background speakers.⏎Our Faculty Members⏎Mr John Norberry: email@example.com⏎Mrs Laura Wade: firstname.lastname@example.org⏎Quite often as Language teachers, we are asked why students should have to
-
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
-
give you initial support and signpost you to appropriate help if required.⏎If you need help or have any questions about Mental Health First Aid please email: firstname.lastname@example.org or talk to your tutor.⏎Are you
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: what is the most significant problem in the world? What is the most significant problem in the world? I think that it's possible to have a lot of people's questions, but I don't think that it's possible to have a lot
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
s, families in the top 20 percent of income continued to see steady income gains, whiles families in the lower income brackets saw stagnant and/or falling incomes.⏎Why? Just
-
extent of the past forest line and the corresponding minimum shift in mean July temperatures, compared with the present.⏎The mean July temperatures in Finnish Lapland have been ca.⏎I have also had restricted access to my extensive holiday wardrobe, thanks to my pes
-
(London, 1900); Macallum, On the Detection and Localization of Phosphorus in Animal and Vegetable Cells, Proc. Roy.⏎A curious fact, which may be used for the detection of the minutest quantity of
-
, what types of cells they become, and how to control those complex mechanisms to facilitate development of new therapies.⏎Researchers a the Burnham Institute for Medical Research (Burnham) and The Scripps Research Institute
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
- L2:2005 suppresses
- L2:237 excites
- L2:111 weakly suppresses
- L0:1301 weakly suppresses
- L1:1464 weakly excites
- L2:394 weakly excites '@example' in email addresses
- L2:424 weakly excites
- L2:2196 weakly excites
- L2:1321 weakly suppresses
- L2:70 weakly excites
- L0:512 weakly suppresses
- L2:122 weakly excites
- L2:49 weakly suppresses
- L1:1029 weakly excites
- L2:1710 weakly suppresses
- L0:178 weakly suppresses
- L2:45 weakly excites
- L2:1533 weakly excites
- L0:1224 weakly suppresses
- L2:527 weakly excites
- L1:146 weakly excites
- L1:2850 weakly suppresses
- L2:493 weakly suppresses
- L2:3070 weakly excites
Feeds later layers
- L4:89 strongly suppresses
- L8:1515 suppresses
- L4:1044 suppresses
- L4:1262 suppresses
- L8:1214 suppresses
- L15:2885 suppresses
- L10:258 suppresses
- L9:935 suppresses
- L4:735 suppresses
- L23:194 weakly suppresses
- L21:1150 weakly excites
- L4:675 weakly excites
- L5:2804 weakly suppresses
- L8:967 weakly suppresses
- L6:324 weakly suppresses
- L12:1435 weakly suppresses
- L12:2508 weakly suppresses
- L6:2644 weakly suppresses
- L24:96 weakly suppresses
- L5:381 weakly suppresses
- L10:762 weakly suppresses
- L16:2255 weakly suppresses
- L27:953 weakly suppresses
- L12:1746 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
- erne
- #⏎⏎
- 寡
- orse
- /Gate
- ledge
- +⏎⏎
- (),⏎⏎
- 族自治县
- ",⏎⏎
Suppresses
- its
- itself
- ewan
- udiant
- ��作
- ��
- itin
- ifice
- tut
- ぶり