'going' or 'go' after a noun or 'is'
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. 20 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 words such as 'going', 'go' and 'true' that follow a subject or a form of 'is', as in 'the ones who are going to live' or 'courage to go'. The snippets are loosely related, and turning the neuron up gave text that did not clearly connect to the prompts.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
another smash hit would break through in American entertainment. Film director Spike Lee’s “Do The Right Thing” was set in the same multicultural Brooklyn as “The Cosby Show,” but
-
failure. Think of it as time-released success.⏎21. There are no failures. Just experiences and your reactions to them.⏎22. Courage is going from failure to failure without losing enthusiasm.⏎23. Failure is unimportant.
-
most of the range and the crests of the adjacent San Gabriel range have lost their kiss of snow, big San Gorgonio often still shines white—the one true jewel
-
are paying attention to what is happening at the COP16 climate negotiations in Cancun, Mexico. They are the “climate generation” — the ones who are going to have to live with our climate legacy, our tendency to prevaricate and
-
fearful aspects of being small in a big world. Bravery is having the courage to go out every day and tackle what comes, no matter what comes. These creatively structured
-
Discoveries class on code.org, jump into CircuitPython to learn Python and hardware together, TinyGO, or even use the Arduino IDE. Circuit Playground Express is the newest and best Circuit
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: It is possible to show that the same between the same is to show that between the same is to show that between the same is to show that between the same is to show that between the same is to show that between the same is to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and air conditioning systems, it also means that pollutants that originate indoors are getting trapped inside.⏎“People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落
-
221BC). Unlike most Chinese idioms, 毋忘在莒 strangely has actually two stories of origin, which gave this idiom two
-
Nihon Shoki 日本書紀 (720) 9 and in the Hizen-no-kuni fudoki 肥前國風土記 (first part of the 8th century) 10. The
-
become the Duke of Qi and started building up the state with reform and dedication. Years passed on since his days as a helpless young heir in the state of Ju, he started to really enjoy the good old
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
- L18:125 suppresses
- L15:59 suppresses
- L11:223 weakly excites
- L19:28 weakly suppresses
- L2:49 weakly suppresses
- L18:66 weakly suppresses
- L10:155 weakly excites
- L4:2662 weakly suppresses
- L16:1308 weakly excites
- L8:328 weakly excites
- L12:2243 weakly suppresses
- L16:513 weakly excites
- L4:607 weakly suppresses
- L13:44 weakly suppresses inside web addresses and citations
- L4:16 weakly excites
- L12:1628 weakly excites
- L1:2922 weakly excites
- L5:2115 weakly excites
- L7:2217 weakly suppresses
- L13:1848 weakly excites
- L16:2225 weakly suppresses
- L9:1129 weakly excites
- L18:1156 weakly excites
- L0:1697 weakly suppresses
Feeds later layers
- L24:1920 weakly suppresses
- L26:2449 weakly excites
- L22:2073 weakly excites
- L22:1075 weakly suppresses
- L23:582 weakly suppresses
- L24:320 weakly excites
- L22:2900 weakly suppresses
- L21:2041 weakly excites
- L24:1650 weakly suppresses
- L22:2619 weakly excites
- L24:2584 weakly excites 'in' in set phrases like 'in mind'
- L21:1652 weakly suppresses
- L25:2255 weakly excites
- L21:2856 weakly excites
- L21:1267 weakly suppresses
- L25:1725 weakly excites
- L25:1354 weakly suppresses
- L21:1034 weakly excites
- L23:1994 weakly excites
- L25:2056 weakly suppresses
- L25:2991 weakly excites
- L21:2670 weakly excites
- L23:1626 weakly suppresses
- L26:787 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
- plete
- ative
- dictions
- mnop
- tega
- urrence
- utive
- like
- iko
- missive
Suppresses
- 之
- 之間
- 之间的
- 之人
- 之间
- 起来
- 起來
- arasında
- 之乡
- 之内