non-English text, mid-phrase
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. 23 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 mostly fires inside sentences written in French, Spanish or other non-English languages, often on a piece of a longer word. Its effect on the written output was hard to read: the steered text drifted between topics and languages with no clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
roof generates more solar power than the school consumes, especially in the summer months and during weekends. The NPO APERe (Association pour la Promotion des Energies Renouvables) and Brussels grid operator Sibelga selected the school as a
-
Museum of Alberta, Musée national des arts et traditions populaires, Département de l'organisation des systèmes d'information
-
this milestone overtime because of our rigorous hiring process that filters professionals from average writers. We believe that exposing our customers to qualified writers is<|endoftext|>Investigación con células madre embrionarias humanas revela los primeros pasos en el desarrollo
-
adie pour 2014 – Rapport au Ministre chargé de la Sécurité Sociale et au Parlement sur l’évolution des charges et produits de l’Assurance Maladie au titre de 20
-
In the following example, hijos is a masculine, plural noun so the masculine plural direct object pronoun los is used to replace it in the second sentence.Son muy buenos hijos. No los cambiaría.⏎They are good
-
ics.org/incentives/020828-01.htm.⏎http://www.serna.gob.hn/ (secretaria de recursos naturales y de ambiente)<|endoftext|>SCAMwatch is aware that some consumers have
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: it's always so easy to set it as the only one, which is known as "first in" or "last in". It's also known as "active" or "active as it is". What's the difference between
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
975) Theory of Reasoned Action (TRA) and Ajzens (Ajzen 1985) extended Theory of Planned Behavior (TPB) model. The TRA
-
85 CE), Japanese art flourished, and new forms of art, such as screen painting, began to emerge. The Kamakura period (1185 CE to 1333 CE) saw the rise of Zen Buddhism
-
frustrated and lose interest in it when it doesn’t show immediate commercial applications. I read just recently about how the science minister of Canada made statements about how the National Research Council isn’t going to pursue projects with immediate commercial applications and it makes me sad
-
To view this ebook on an iPhone, iPad or Android mobile device you will need the Adobe Digital Editions app, or BlueFire Reader or Txtr app. These are free, too. For more details see this article.⏎|Size: ||
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
- L19:1422 strongly excites
- L20:807 strongly excites
- L21:607 excites
- L19:1743 excites
- L19:639 excites
- L20:240 suppresses
- L18:2391 excites
- L19:1424 excites
- L19:1427 suppresses
- L20:1842 suppresses
- L21:1376 suppresses
- L21:266 excites
- L8:253 suppresses
- L19:1695 excites
- L17:1583 excites
- L7:327 excites
- L9:1062 excites
- L18:444 suppresses
- L16:1418 excites
- L14:1390 excites
- L18:1433 excites
- L19:2131 excites
- L19:2382 suppresses
- L19:1909 excites
Feeds later layers
- L24:434 strongly excites
- L27:2186 strongly excites
- L25:335 strongly excites
- L25:812 strongly excites
- L24:360 suppresses
- L26:111 excites
- L27:60 suppresses
- L25:403 excites
- L24:109 suppresses
- L26:552 excites
- L27:477 excites
- L25:377 excites
- L24:419 excites
- L27:1010 excites
- L25:320 suppresses
- L24:489 suppresses
- L26:441 suppresses
- L27:1052 excites
- L25:464 suppresses
- L26:1017 excites
- L26:437 suppresses
- L27:227 excites
- L24:488 excites
- L27:307 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
- 們
- 家长们
- ackers
- majors
- 们
- ones
- 学生们
- targets
- 老师们
- (targets
Suppresses
- который
- щий
- ющая
- ющий
- которое
- Its
- метр
- енный
- ístico
- Its