"how" after learn or know
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. 18 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" when it follows verbs like "learn", "know" or "determine", as in "learn how to play". Turning it up did not clearly bring out that theme in the sample writing, though the examples are very consistent.
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.
-
tell that an essay has been plagiarized, even if unsure of the exact source of pirated passages. Plagiarism software can be used to help determine how original - or unoriginal - a student's written work may be. A student
-
listening to your favorite pieces -- a real extra dimension for your music enjoyment.⏎Later you will also be able to use your ability to read music notes if you learn how to play a musical instrument.⏎How to Read Music Notes - Introduction⏎No comments
-
them to new textures and sights during their tummy time.⏎You can place their favorite toy in front of them so that they can practice their motor skills and learn how to grab.⏎Tummy time fosters independence and a sense of learning in a
-
grade children should be giving consideration and paying more attention to the correct spelling of words. They should be spelling previously learned words correctly. If the child does not know how to spell a word, he or she
-
of the misconceptions spread about it by its critics - Angst… that the Energiewende will work.⏎Unlike the Swedes and Italians, the Germans knew how to replace nuclear back in 2002 – and they have even replaced
-
and adopted each year and will penalize laboratories that do not comply with adoption standards. Currently, labs are required to report to the U.S. Department of Agriculture how many animals are used each year, but these reports can be misleading. USDA reports
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 have been able to take some ideas and say a more concrete and more conscious awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness awareness
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’s super easy to get started too.⏎Here’s a quick how-to:⏎- Get in a comfy position. You don’t need to pretzel it up, know lying down can lead to sleeping, but other than that, just be comfortable in
-
believe that we owe this story to those who came before us.⏎As a young person growing up in this country, I remember my parents telling me, “You know what?⏎We have to get out of this country and go to another country,
-
I’m Jane Houston Jones at NASA’s Jet Propulsion Laboratory in Pasadena, California.⏎Have you ever seen the Space Station pass overhead? It’s easy if you know when and where to look.⏎The Space Station orbits Earth about 15 times
-
said Guillaume Gronoff, a French astrophysicist at NASA Langley.⏎"And you think, 'OK, what I am looking at — it's a<|endoftext|>An ultra-Orth
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
- L16:2097 suppresses
- L16:1556 suppresses
- L17:913 excites
- L15:2392 suppresses
- L14:2142 suppresses
- L17:2582 excites
- L11:1469 excites
- L10:1919 suppresses
- L17:1952 excites
- L11:1774 excites
- L14:1720 excites
- L17:2772 suppresses
- L14:462 excites
- L17:1562 suppresses
- L15:2053 excites
- L16:2718 suppresses
- L16:1672 excites
- L15:433 excites
- L14:2460 weakly suppresses
- L13:1311 weakly suppresses
- L13:1776 weakly excites
- L10:254 weakly suppresses
- L17:1321 weakly suppresses
- L9:1453 weakly excites
Feeds later layers
- L19:1631 suppresses
- L19:1808 excites
- L21:918 excites
- L19:1483 suppresses
- L20:2495 suppresses
- L19:1548 excites
- L19:1177 excites
- L19:1824 suppresses
- L20:858 suppresses
- L19:1796 excites
- L19:2143 suppresses
- L22:3009 suppresses
- L19:580 excites
- L23:594 excites
- L22:2067 suppresses
- L19:2643 weakly excites "is" in "what is" questions
- L20:330 weakly excites
- L19:1055 weakly excites
- L22:543 weakly excites
- L19:1396 weakly suppresses
- L21:1910 weakly suppresses
- L20:2433 weakly suppresses
- L19:2117 weakly suppresses
- L20:1413 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
- 为何
- why
- ela
- iele
- why
- how
- 何时
- 為什麼
- يا
- 如何
Suppresses
- 的认识
- 知道
- 的理解
- 描述
- 了解到
- 了解
- 启示
- identified
- asurer
- 理解和