"and" joining questions (wh-question context)
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. 17 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 "and" when it follows a question, joining it to a second question as in "Where will that money come from, and why...". It works in the opposite direction on words like "who" and "what" inside embedded questions. Turning it up made the model repeat the word "what" over and over, which fits its link to question words.
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.
-
guides, fun facts, and photography you find here will inspire you to explore new places!<|endoftext|>Will there be radiation when the mobile phone signal is shielded, and is it harmful to the human body?⏎Regarding radiation, as long as electronic products
-
County, there is now a new birth and a new way to remember a rich history .⏎What happened to the hamlet of Huntertown in Woodford County, and
-
have recently been revised. We explore their history and increase in popularity.<|endoftext|>Native Plants for the Mid-Atlantic⏎On this page⏎- What Are Native Plants and Why Are They Important?⏎- State-Specific Native Plant Organizations⏎- Other Sources
-
mineral or non-mineral based oils) or a synthetic blend (or hydrocarbon-based oils and synthetic-based oils).⏎But which motor oil is best for your vehicle and what’s the difference between all the different types, weights and other options
-
worlds around other stars has shown that planets orbit at least 5 to 10 percent of all stars. But how many of these planets are Earth-size, and possibly Earth-like?<|endoftext|>& Tornado Alley⏎Jan Mikusinski developed an
-
Fred Gibbs and I were awarded from Google to mine their vast collection of books.]⏎Why did the Victorians look to mathematics to achieve certainty, and how we might understand the Victorians better with the mathematical methods they bequeathed
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 can see, and I can know, and I can know more? And I can know more, and I can know more, and I can know more? The more I know, the more I can know? This is a part
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
%. In order to achieve this objective the marketer has to prepare a plan in respect of the level of production and promotion efforts. It will also be decided as to who will do what, when and how. To do this is known as marketing planning
-
needed. While it may sound trivial, the decision on what to produce/grow has important ramifications. If you plan to grow/produce a
-
battery. This article will be broken into two parts, the "How-to" and the "Why." The How-to section features detailed instructions with illustrations, including what tools you'll need, how to wire connections, and other important safety tips.
-
at home. Here are some simple tips that you can make your garden look more beautiful.⏎According to expert, before making a garden, first you have to consider what is the concept of the building of your home. After that, adjust your garden
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
- L15:1473 strongly excites
- L11:296 suppresses
- L4:470 suppresses
- L4:927 excites
- L9:75 excites 'what' or 'how' after see or understand
- L8:1997 excites
- L16:2975 suppresses
- L4:2558 excites
- L12:1667 excites
- L5:2657 suppresses
- L9:524 excites
- L5:458 excites
- L13:344 suppresses
- L5:1800 weakly suppresses
- L7:1081 weakly suppresses
- L13:2728 weakly suppresses
- L1:2951 weakly suppresses
- L2:1640 weakly excites
- L0:1878 weakly suppresses
- L16:817 weakly suppresses
- L6:687 weakly excites
- L0:1966 weakly suppresses
- L5:702 weakly suppresses
- L10:722 weakly excites
Feeds later layers
- L21:728 strongly excites
- L18:2517 excites
- L23:594 excites
- L27:656 excites
- L19:2643 excites "is" in "what is" questions
- L19:2814 excites
- L18:2031 suppresses
- L25:576 excites
- L20:1371 suppresses comma inside a question
- L22:543 excites
- L18:1586 excites
- L25:541 suppresses
- L26:1046 excites
- L19:1703 suppresses
- L24:1439 excites
- L27:233 suppresses
- L22:1933 excites
- L26:400 excites
- L19:1514 weakly excites
- L22:1414 weakly excites
- L26:1836 weakly excites
- L23:1089 weakly excites
- L26:1104 weakly excites
- L18:1918 weakly 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
- 为何
- 如何
- 为什么
- why
- how
- 多么
- 何时
- 什么原因
- 什么意思
- 有多少
Suppresses
- ?
- ?\
- 呢
- ? ⏎
- ?
- ?⏎
- ?p
- ?⏎⏎
- ?,
- ef