line breaks before lists or answer options
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. 4 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 line break right after a sentence that introduces a list, and on capital letters that start answer choices like 'A' or 'D' in quizzes. It goes negative on the word 'RV' or 'home' inside RV and camper articles. Turning it up gave random quiz-like text, which fits the quiz side but is not conclusive.
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.
-
(1), they were loyal because ‘others’ told them to be loyal (2), and they could not be disloyal to the British as the sol⏎The historical literature on Afghanistan and the various armed conflicts fought on its soil has greatly
-
result, the body gets amino acids that consist of protein molecules.After splitting the latter formed a toxic substance - ammonia.It brings the bloodstream to the liver, where b⏎Why do we determine the content of urea in the blood?⏎What is going
-
College, and has delivered guest lectures at Yale University, SUNY Purchase, and West Chester University.<|endoftext|>There's the Old English dumb, but for⏎DUMB the dictionary offers the Indo-European root dheu (to rise in a cloud
-
areas such as assertiveness, communication, relationships, and self-esteem One of the main goals of feminist therapists is to develop equal mutual relationships of caring and suppor⏎Who created feminist theory?⏎Charlotte Perkins Gilman’s⏎What is feminist intervention?⏎
-
only on women who are at increased risk of having a fetus with aneuloidy. Among the indications listed for women considered to be at increase risk are:⏎- Maternal age 35 years or older at delivery⏎- Fetal
-
two end systems⏎B. It specifies the quality of the data & connection⏎C. It specify the size of data⏎D. All of the above⏎6-4. The class-based addressing is also
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 car is 2010, and the load is 2010. The car is 2010. The car is 2010. The car is 2010. The car
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’t Try Climbing Mt. Rainier Right Now - September 21, 2021⏎- The Sylvansport Electric RV is the Motorhome of Our Dreams - September 14, 2021⏎-
-
need, how to wire connections, and other important safety tips. This blog post's Why section explains why RV owners should be able to jump-start their camper or RV battery in an emergency.⏎Table of Contents⏎- What Is a Battery and
-
types of batteries.⏎Lead Acid Battery: They take longer to lose their charge when not used, making it easier to maintain your battery while away from the camper or RV.⏎Nickel Cadmium Battery: These batteries have been around a long time and
-
urgeon Study⏎- Townet Survey⏎- Zooplankton Study⏎- All Surveys and Studies⏎Other BDR Links⏎7329 Silverado Trail⏎Napa, CA 94558⏎210
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
- L2:111 suppresses
- L3:894 weakly suppresses
- L3:310 weakly suppresses
- L1:929 weakly excites
- L1:64 weakly suppresses
- L3:59 weakly excites
- L1:1349 weakly excites
- L3:691 weakly excites
- L0:2168 weakly excites
- L2:1358 weakly suppresses
- L1:1666 weakly suppresses
- L3:272 weakly excites
- L2:2410 weakly excites
- L2:1109 weakly suppresses
- L2:2912 weakly suppresses
- L2:1334 weakly suppresses
- L1:2340 weakly suppresses
- L3:1303 weakly suppresses
- L2:2326 weakly suppresses gravity word pieces (gravit-)
- L1:1097 weakly excites
- L3:1937 weakly excites
- L2:2385 weakly excites
- L3:1037 weakly suppresses
- L3:768 weakly suppresses
Feeds later layers
- L9:1328 weakly excites
- L5:749 weakly excites
- L9:1732 weakly suppresses
- L7:990 weakly excites
- L11:1021 weakly suppresses
- L5:375 weakly excites
- L7:1625 weakly excites
- L13:1560 weakly suppresses
- L22:1643 weakly suppresses
- L8:1351 weakly suppresses
- L5:1977 weakly suppresses
- L7:2990 weakly excites
- L10:2490 weakly suppresses
- L8:1326 weakly excites
- L11:2164 weakly suppresses
- L15:1171 weakly excites
- L15:1237 weakly excites
- L16:410 weakly excites
- L7:1688 weakly excites
- L8:2162 weakly suppresses
- L12:2691 weakly suppresses
- L11:2409 weakly excites
- L19:1926 weakly excites
- L22:2164 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
- quila
- ories
- ORIES
- 閣
- Celsius
- ystore
- stell
- acles
- ctp
- LatLng
Suppresses
- dug
- coon
- ataka
- rek
- rig
- wired
- 万辆
- arded
- _$
- rim