"is" in "what is" questions
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. 19 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 "is" when it comes right after "What" in a question, like "What is the meaning of...?" Its opposite side fires on commas after phrases like "the question is," which points to a shared theme of questions. Turning it down made the model repeat "what" or "how" over and over.
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.
-
There is now firm evidence supporting the lever arm hypothesis for the origin of directional movement. Nevertheless, many pivotal issues remain. For example, what is the sequence of events in the conversion of the chemical energy derived from ATP hydroly
-
poem serves as a reminder of the importance of connecting with nature, finding solace in its embrace, and seeking a deeper understanding of oneself.⏎Q1: What is the meaning behind Mary Oliver’s poems?⏎A1: Mary Oliver’s poems often
-
There⏎are two rows of seats in the waiting room. Using the rules, where⏎are they all sitting? Can you find all the possible ways?⏎What is the best way to shunt these carriages so that each train⏎can continue
-
Hittites?⏎- What is the meaning of Hittites?⏎- What does Itin mean?⏎- What
-
<|endoftext|>What is Cancer? What Causes Cancer?⏎Cancer is a class of diseases characterized by out
-
.What is the difference between MOV and MVI?⏎12. What are the functions of
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: How can what what how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how how
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, and there is not much chance of getting the genie back in the bottle. It is here and you better believe it’s impacting your business. The question is, what are you going to do about it?⏎The term “Green” comes from
-
, the organic acids that are part of each kefir grain readily react with reactive metals. The important question is, which metals are reactive in the process of making and storing kefir? The answer
-
to move beyond simply complaining, reflexively and vaguely, that government is "too big." They must instead develop a concrete, practical approach to answering the question, "What do you propose to cut?" — a realistic set of guidelines that elucidate
-
will open two charter schools in Baton Rouge. Like many parents, you may be wondering what exactly is a charter school? And more importantly, is it a good educational option for my child?⏎What is a Charter School?⏎
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
- L17:2638 strongly suppresses
- L17:1480 strongly excites
- L16:2995 strongly excites
- L18:1586 suppresses
- L18:2010 excites
- L17:1681 excites
- L17:1407 excites "and" joining questions (wh-question context)
- L12:2540 suppresses
- L7:2761 suppresses
- L18:2562 excites
- L15:1871 excites
- L7:2058 excites
- L18:1502 suppresses
- L14:1800 suppresses
- L18:2666 suppresses
- L17:1952 weakly excites
- L16:1942 weakly suppresses
- L16:2097 weakly suppresses
- L1:1464 weakly excites
- L18:3053 weakly excites "how" after learn or know
- L13:2728 weakly suppresses
- L16:2204 weakly suppresses
- L8:2326 weakly excites
- L14:492 weakly suppresses
Feeds later layers
- L27:656 strongly suppresses
- L21:728 strongly suppresses
- L22:1414 strongly suppresses
- L23:594 suppresses
- L22:543 suppresses
- L22:2255 suppresses
- L26:1836 suppresses
- L26:400 suppresses
- L23:1089 suppresses
- L25:576 suppresses
- L26:1046 suppresses
- L20:1043 suppresses
- L24:1439 suppresses
- L22:1592 suppresses
- L25:1407 excites 'in' right after a place or time
- L21:360 suppresses 'such' introducing examples
- L26:1887 suppresses
- L20:1114 suppresses
- L25:2039 suppresses
- L23:197 weakly suppresses
- L23:2187 weakly suppresses
- L27:741 weakly suppresses
- L26:1104 weakly suppresses
- L20:73 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
- ỗ
- vine
- iminal
- cular
- cgi
- 我认为
- ıyla
- azard
- ucas
- amment
Suppresses
- how
- how
- 如何
- 哪些
- 怎样
- what
- 怎么
- How
- 怎麼
- 哪