"how" after "learn" or "demonstrate"
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. 6 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 word "how" after phrases like "you will learn how" or "demonstrate how", where a how-to explanation is about to follow. Turning it up or down did not produce a clearly related change in the written text.
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.
-
and clear. Deciding upon the structure of a report can often be one of the most challenging tasks, particularly for un<|endoftext|>In this tutorial, you will learn how to illustrate a bee icon using simple shapes and colors. This adorable insect can be
-
textbook will demonstrate how an answer is reached. Students can use the textbook to find answers and explanations.
-
majority of cancers, may involve an interplay between specific environmental factors related to epigenetic regulation of the genetic blueprint. Epigenetics provides a compelling model for how environmental influences throughout development can have life-long (and potentially multi-generational)
-
inders where the child must find the matching sound pairs by listening and then grade the boxes from the loudest to the soft<|endoftext|>In this tutorial, you will learn how to animate a bouncing ball lazily and really quickly!⏎If you plan to become
-
at innovative mathematics.⏎Numbers worksheets enable you to be creative inside your training of the little one. They motivate your child to participate in within the training and show the way they learn by being more productive within
-
tables and columns as well as calculating summary statistics such as means, frequency, and percentiles. Creating filters and joins on in-memory data are also discussed.Learn how to⏎Who should attendExperienced programmers who need to learn the syntax and functionality
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 just want to know the truth. For the time being, I just need to know if this is the best answer. Let me know if I can get it. So what's the answer. Now I need to know if this is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
The effects of acid precipitation on the ecological environment and restoration. Press of Chinese Environmental Sciences, Beijing, pp 1–3⏎Lei HC, Wu YX, Xiao H, Shen ZL (2001) Study on cloud
-
⏎Lin CC, Lin XM, Zou Y, Zhang L (2005) Study on the
-
four different arcs, for a total of twelve separate images of the same galaxy!⏎Whoa. The cluster is called PSZ1 G311.65−18.48, and
-
to a long-term and repeated drought in monoculture and mixed plantings, the “Kranzberg Roof Project” (KROOF, kroof.wzw.tum.de) was initiated in 2013 and received funding by
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
- L4:1651 suppresses
- L3:213 weakly suppresses
- L4:532 weakly suppresses
- L3:1383 weakly suppresses
- L5:66 weakly excites
- L3:893 weakly suppresses
- L3:156 weakly excites
- L5:2256 weakly suppresses
- L1:2306 weakly excites
- L0:3061 weakly excites
- L5:1857 weakly excites
- L5:1900 weakly excites
- L3:2218 weakly suppresses
- L4:2240 weakly excites
- L4:2848 weakly suppresses
- L3:3042 weakly excites
- L1:2350 weakly excites 'reasons' or 'permitting' after a qualifier
- L4:12 weakly suppresses
- L1:1992 weakly suppresses
- L4:1075 weakly suppresses
- L2:111 weakly suppresses
- L4:868 weakly excites
- L3:1003 weakly excites
- L2:3005 weakly excites
Feeds later layers
- L7:55 suppresses
- L13:645 weakly excites
- L17:758 weakly suppresses
- L23:2158 weakly excites
- L9:207 weakly suppresses
- L8:2244 weakly suppresses
- L24:838 weakly excites
- L8:200 weakly excites
- L18:2830 weakly excites
- L10:370 weakly excites
- L20:395 weakly suppresses
- L9:232 weakly suppresses
- L8:277 weakly excites
- L10:1159 weakly excites
- L9:281 weakly suppresses
- L10:2432 weakly excites
- L15:401 weakly excites
- L21:2400 weakly excites
- L14:1818 weakly excites
- L24:1165 weakly excites
- L17:435 weakly suppresses
- L24:1268 weakly excites
- L14:3003 weakly suppresses
- L16:1627 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
- entials
- 郦
- .shuffle
- gether
- 医科大学
- panion
- 匏
- strength
- rv
- 鹬
Suppresses
- ē
- 台
- اص
- 节
- 较大的
- Dön
- agements
- agen
- chù
- [