line breaks before page-section headings
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 invisible line break that comes right before a heading or a menu item on a web page, such as 'Back to top' followed by a section title. Turning it up made the text get stuck in loops, so the link to web-page layout is only a rough guide.
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.
-
from you. Maybe they never let you off the hook, even if you got 98 out of 100 on an exam.⏎Back to top⏎The Role of Indirect Aggression⏎What goes on outside the home is also
-
West.<|endoftext|>Two dimensional spatial coherence of skeletal muscle's natural vibrations during voluntary contractions.⏎Archer, Akibi A. A.⏎MetadataShow full item record⏎Low frequency mechanical vibrations (<100 Hz) are naturally generated by skeletal muscles
-
, have of you⏎- Having high expectations of other people⏎But whatever form it takes, perfectionism can rob you of life's pleasures.⏎Back to top⏎The Roots of Perfectionism⏎It may not be so easy to figure out
-
06,<|endoftext|>Book/Printed Material Kings of Arabia; the rise and set of the Turkish sovranty in the Arabian peninsula,⏎About this Item⏎- Kings of Arabia; the rise and set of the Turkish sovranty
-
of balloons.⏎Submit a review<|endoftext|>1 year ago⏎Verified Sherpa Tutor⏎If a liquid has a definite volume, what this would mean is that you can
-
the persecutions suffered by Jews and Judaism in exile was not the decimation of our numbers, but the injury to Jewish...⏎Become a Member⏎Read all member content⏎Get exclusive in-depth reports from Israel.⏎Get exclusive in
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 I E the in_ eInI1_tT__nT T__In_In_In_Pa_In_H_In_In_P_p_hStartH_pHi in_pP_start
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Choose the book you like when you register 4.⏎Discussion questions for ‘The Death and Life of the Great Lakes’⏎Skip to search form Skip to main content You are currently offline. Some features of the site may not work correctly.
-
rairie View A&M College⏎Rights© 2021 Prairie View A & M University⏎This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.⏎Date of Digitization⏎John B Coleman Library
-
NIZZOLI, Giampaolo ROSSETTI, Pierluigi VIAROLI⏎This work is licensed under a Creative Commons Attribution-NonCommercial 4.
-
⏎Copyrights and publishing rights of all the texts on this journal belong to the respective authors without restrictions.⏎This journal is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License (full legal code).⏎
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
- L1:316 excites
- L0:1449 weakly suppresses
- L1:1623 weakly suppresses
- L1:1743 weakly suppresses
- L1:2340 weakly suppresses
- L1:1721 weakly suppresses
- L1:2683 weakly excites
- L0:1547 weakly suppresses
- L1:572 weakly excites
- L1:1464 weakly excites
- L1:2172 weakly suppresses
- L1:917 weakly excites
- L0:299 weakly excites
- L0:1265 weakly excites
- L1:1340 weakly suppresses
- L1:1407 weakly excites
- L1:1099 weakly excites
- L1:416 weakly suppresses
- L1:1651 weakly suppresses 'of' after 'full'
- L1:595 weakly excites
- L1:1126 weakly excites
- L1:251 weakly excites
- L0:479 weakly suppresses
- L0:1369 weakly suppresses
Feeds later layers
- L3:115 suppresses
- L3:121 suppresses
- L4:1188 suppresses
- L13:151 weakly excites
- L4:245 weakly suppresses
- L3:767 weakly suppresses
- L19:2452 weakly suppresses
- L4:38 weakly suppresses
- L4:2521 weakly suppresses
- L8:432 weakly suppresses
- L12:1240 weakly suppresses
- L5:2813 weakly excites 'in place' after 'in'
- L6:1161 weakly suppresses
- L7:997 weakly suppresses
- L14:789 weakly excites
- L3:2926 weakly suppresses
- L3:1161 weakly suppresses first digit after 'more than'
- L13:2982 weakly suppresses
- L10:1952 weakly suppresses
- L3:945 weakly suppresses
- L6:2437 weakly suppresses
- L17:1323 weakly suppresses
- L3:480 weakly suppresses
- L3:2578 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
- ственные
- agrams
- disciplinary
- تك
- into
- grids
- aturas
- agna
- indeed
- Guinea
Suppresses
- center
- http
- OOT
- ocation
- https
- endment
- cent
- Third
- former
- oz