'to' after a noun or verb of motion
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 "to" when it comes right after a noun or verb that is followed by an infinitive, such as "helicopter to scope" or "going from failure to failure". Turned up, the model's replies drifted into odd, unrelated topics.
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.
-
agricultural workers what realities of their lives you found most disturbing about their working conditions. Please be specific and cite your source.Please identify two reasons agricultural workers leave their homeland to work in the fields of American agricultural for low pay. Is it possible that free
-
) Before humans make it to Mars, NASA will send a helicopter to scope out the terrain.⏎The solar-powered Mars Helicopter will be safely stowed
-
as Pyrrhus the Delian fisherman, and then Pythagoras.There is a famous story<|endoftext|>Have you ever gone through the harrowing experience of going to grab some milk, wanting to know how much was
-
worry about failure.⏎31. A true friend is one who overlooks your failures and tolerates your success!⏎32. Success consists of going from failure to failure without
-
both rescue the continuity and advance the scientific capability of the mission, if it reaches orbit. If it doesn’t, it won’t be the first Landsat satellite to fail.⏎In 1993, Landsat 6 didn’t make
-
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. It takes courage
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: the only way to make it work, if it was a problem that was fixed by the problem of the other’s problems. Which of the following is the most likely best friend for a problem problem problem? A) someone who is the only
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
beekeepers use. However, the bees usually win out. Once the queen cells are eliminated the bees will produce more queen celIs.⏎|Prepared by:||Gordon Rudloff⏎Ohio Department of Agriculture<|endoftext|>Logic & ProofsOpen Learning
-
crosses directly through the desert with the Great Bitter Lake bein part of the canal. In the background the Sinai Peninsula and the Red Sea.⏎|View:||The Suez Canal|<|endoftext|>Old Times literature essays are academic essays for citation.
-
grow effectively in water:-⏎Ways to care for Monstera and make them grow faster:-⏎The Monsteras are widely growing plant types. They have a lot of ways to
-
as under :- 1. Introduction :- Mediation introduces the parties, explains mediation process, establishes neutrality and eventually generates momentum towards settlement. 2. Joint Session :- Mediation gathers information about the factual background and interests of the parties. Establishes
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
- L14:348 suppresses
- L10:33 suppresses
- L9:7 excites
- L11:18 excites
- L15:44 suppresses
- L5:5 suppresses digit after a label like paper or type
- L11:31 suppresses
- L13:132 excites "of" after a noun
- L16:170 suppresses
- L14:51 excites colon that introduces a statement
- L15:123 suppresses
- L14:917 excites
- L14:303 excites
- L3:3 suppresses
- L1:1 excites
- L16:215 suppresses
- L13:1046 suppresses 'in' or 'on' ending a phrase
- L12:5 excites
- L14:251 suppresses
- L6:4 suppresses
- L16:337 excites
- L16:1596 suppresses
- L1:8 excites
- L15:798 weakly excites
Feeds later layers
- L19:315 suppresses
- L19:926 suppresses
- L19:380 suppresses
- L21:122 suppresses
- L18:655 suppresses
- L20:129 suppresses
- L18:777 suppresses
- L20:247 suppresses
- L19:183 suppresses
- L18:467 suppresses
- L18:459 suppresses
- L20:463 suppresses
- L19:402 suppresses
- L20:133 suppresses
- L18:2498 suppresses
- L18:285 suppresses
- L20:190 suppresses
- L20:119 suppresses
- L19:308 weakly suppresses
- L18:1186 weakly suppresses
- L25:535 weakly excites commas inside number lists
- L19:283 weakly suppresses
- L20:197 weakly suppresses
- L19:243 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
- チョ
- 俟
- dden
- ellij
- mö
- ivirus
- @end
- spd
- झ
- .idea
Suppresses
- igrams
- ang
- ⏎⏎⏎
- ather
- ⏎ ⏎
- -grand
- omb
- ⏎
- itr
- ⏎⏎