words inside headlines and titles
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. 13 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 words inside short headlines or article titles, usually the word right after the first few words of a headline, such as 'Ketamine [Found] Effective' or 'Survey [Reve]als'. Turning it up did not produce a clear headline style in the output.
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.
-
Pesticides in Produce⏎Researchers Highlight Pesticides in Produce,⏎Baby Food, Tap Water⏎Washington, D.C
-
a member of Curiosity's navigation team, told SPACE.com. "It was just like clockwork, the spacecraft behaved so well."⏎- Curiosity Camera Snaps Heat Shield Separation | Video⏎- Mars Rover Curiosity: Mars Science Lab
-
according to the U.S. Drug Enforcement Administration (DEA).⏎Ketamine Found Effective, but More Study Suggested⏎The researchers found that ketamine significantly reduced
-
& Drug Administration approval, and found that those states that skewed toward newer drugs tended to have above-average increases in longevity, when adjusted for other factors.⏎Multiple Factors at Work⏎In New York, for example, the average drug approval year for patients
-
amil, or Mona S. Rowe⏎FOR IMMEDIATE RELEASE⏎DECEMBER 3, 1997⏎Images of Hypericin Molecule, St. Johns Wort flower & Ed Castner, BNL chemist and co
-
about it<|endoftext|>Press Release 07-030⏎Survey Reveals Family Ties and Traditional Activities Keep Arctic Communities Vital⏎Results will be released
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: A. A. A. A. A. A. A. A. A. A. A. A. A. A. A. A.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
letters: C F O S U.⏎No direct anagrams for focus found in this word list.⏎List shorter words within focus, sorted by length⏎All words formed from focus by changing one letter⏎Browse words starting with focus by next letter<|endoftext|>
-
immigrants' rights groups to protest legislation in California that would criminalize delivery of public health, education and welfare services to undocumented immigrants.⏎by Ched Myers⏎All articles on this site were written by Ched Myers unless otherwise specified.<|endoftext|>With Thanksgiving
-
help ‘rewire’ the brains of people with depression, offering hope for patients whose condition is not improved by medication or therapy.⏎Share this article⏎Depression is one of the most common forms of mental illness, affecting
-
Through your library|⏎References found in this work BETA⏎No references found.⏎Citations of this work BETA⏎No citations found.⏎Similar books and articles⏎Piero Antuono & Jan Beyer (19
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
- L11:45 suppresses
- L10:471 excites
- L10:270 suppresses
- L4:887 suppresses
- L11:784 suppresses
- L8:311 excites
- L7:728 excites
- L12:1310 suppresses
- L9:173 weakly excites
- L10:2376 weakly suppresses
- L9:122 weakly excites
- L10:1516 weakly excites
- L0:1145 weakly excites
- L12:605 weakly excites
- L12:982 weakly excites
- L6:234 weakly suppresses
- L3:14 weakly suppresses
- L9:2447 weakly excites
- L7:216 weakly excites
- L10:58 weakly suppresses
- L10:494 weakly suppresses
- L12:2494 weakly suppresses
- L4:30 weakly excites
- L2:189 weakly suppresses
Feeds later layers
- L17:856 strongly suppresses
- L16:1047 strongly excites
- L18:830 strongly suppresses opening quote marks before a coined term
- L15:1565 strongly suppresses
- L17:743 strongly suppresses
- L17:891 suppresses
- L14:853 suppresses
- L17:114 excites
- L15:667 suppresses
- L15:964 suppresses
- L16:1539 suppresses
- L16:1463 suppresses
- L16:1264 suppresses
- L15:1865 suppresses
- L16:2398 suppresses
- L18:1498 suppresses
- L15:2591 suppresses
- L15:2532 suppresses
- L14:812 suppresses
- L16:1652 suppresses
- L21:194 excites
- L19:260 suppresses
- L14:2061 excites
- L19:389 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
- ."),
- .'); ⏎
- ."),⏎
- .'),⏎
- ."); ⏎
- ,");⏎
- .');
- .").
- ..."); ⏎
- .', ⏎
Suppresses
- :
- ⏎⏎
- $⏎⏎
- Now
- :
- End
- $⏎⏎
- Now
- -