prepositions in fixed phrases
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. 24 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 lights up on the small word that completes a set phrase, like 'under a microscope', 'behind the wheel' or 'in front of a bright light'. Steering it up produced repetitive, broken text, so the effect on writing is hard to read.
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.
-
. By starting earlier it finishes its flow earlier. Thus solves the problem of too much melatonin in the morning. For some people the idea of sitting in front of a bright light first thing in the morning is unpleasant and simply not practical. There
-
the only thing we do that makes us all the same. Motor vehicles are the true equalizer. Remember driving is a privilege not a right, when you’re behind the wheel you are responsible for a 4000 lb. potential weapon
-
A dive, use a submersible vessel, or study marine life from shore. A marine biologist may work in a laboratory, where they may examine tiny creatures under a microscope, on a large research vessel
-
to take action and remove the clear and present danger of mobile phones on our roads.”⏎Have your say | Should drivers be banned from using hands-free mobile phones when behind the wheel? #RSGB— Road Safety GB (@Road_Safety_GB)
-
doctor will ask about your symptoms and medical history. A physical exam will be done. Tests may include:⏎- A sample of the breast discharge to look at under a microscope⏎- Blood<|endoftext|>Welcome to the Computer Science Program!⏎Tarleton State University
-
often perform tissue biopsy (a scientific procedure for collecting cell samples for closer examination) when a patient reports suspicious lump or symptoms.⏎Through examining the appearance of cells under a microscope, the presence of cancer can be determined. This process reveals the type 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 to support, not just to see, but to see, and to see how to see, to see how to see—so that we can see, not just to see, but to see, to see, to see, to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
8 to April 26. What about solar power? On April 19, the unveiling of the solar classroom provided by Con Edison is sure to be eye-opening. And who doesn’t want to see those marvelous birds of
-
"Black-haired and dark-skinned men and women were powdered a ghostly white as they swept the streets for a last fistful."<|endoftext|>Editor's note: Brain Basics from Scientific American Mind is a series of short video primers on the brain
-
ite woman and has children with her.⏎Now, hereís a hint as to why Moses included this in Genesis 3<|endoftext|>Sujata Gupta for Nature – Brain food: Clever eating:⏎Our ancestors’ consumption of meat is what made us
-
a million displaced. It generated an international response that was probably unprecedented in scale. As someone who regularly reads articles on underfunded international crisis appeals, I was heartened by the response to the tsunami. That it hit easily accessible coastal regions,
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
- L5:2652 weakly excites
- L4:2299 weakly excites
- L13:239 weakly suppresses
- L16:395 weakly suppresses
- L0:2570 weakly suppresses
- L7:557 weakly excites
- L21:814 weakly excites
- L22:2518 weakly suppresses
- L7:2724 weakly excites
- L19:1241 weakly suppresses
- L2:3070 weakly suppresses
- L17:2822 weakly suppresses
- L10:3006 weakly excites
- L22:373 weakly suppresses
- L8:1004 weakly excites
- L5:510 weakly excites
- L0:2454 weakly suppresses
- L22:143 weakly excites
- L1:977 weakly suppresses
- L6:576 weakly excites
- L5:1197 weakly excites
- L6:615 weakly excites
- L3:312 weakly suppresses
- L6:2465 weakly excites
Feeds later layers
- L26:507 excites
- L26:2782 excites
- L27:127 weakly suppresses
- L27:797 weakly excites
- L27:425 weakly excites
- L25:490 weakly suppresses
- L27:792 weakly excites
- L25:336 weakly excites
- L25:750 weakly suppresses
- L26:1310 weakly suppresses
- L27:2084 weakly excites
- L27:1486 weakly suppresses
- L26:851 weakly suppresses
- L25:1586 weakly excites
- L25:530 weakly excites
- L25:2569 weakly excites
- L25:906 weakly suppresses
- L25:1335 weakly excites
- L25:1131 weakly excites
- L25:2910 weakly suppresses
- L27:2932 weakly excites capital F starting a word
- L27:1641 weakly excites
- L26:2348 weakly suppresses
- L25:135 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
- stay
- ersen
- ˢ
- umont
- ấm
- ertation
- رام
- The
- trzymał
- oral
Suppresses
- AttribPointer
- idos
- 高等学校
- shoulders
- 照明
- outube
- numbers
- tires
- arily
- less