'to' after surprised or glad
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. 22 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' after a feeling or reaction word, such as 'surprised to', 'pleased to', 'glad to' or 'fascinating to', where a verb like 'see' or 'find' is expected next. It is pushing the model toward words about seeing and reading. When it was turned up, the model wrote loops of 'read' and 'see', which matches that job.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
and easy to understand while being instantly relatable to everyday life.⏎The book ends with a discussion of organic chemistry’s role in nanotechnology. I was especially pleased to see this included as nanotechnology
-
have seen a vaping advert, whether you are a parent with a child in school or an adult who interacts regularly with youth. It’s possible that you were surprised to find out that vaping can lead to addiction and other health problems. But it’s
-
causes of blindness. A couple of years ago my eye doctor told me that I had cataracts. They were not very far along, but I was shocked to have them at the age of 55. My cataracts seem to
-
the shade,” he says. “I try to stay out of the sun.”⏎The study is published in the journal Cancer Causes & Control.⏎“I was quite surprised to find the associations<|endoftext|>|This article needs additional citations for verification
-
use of Read Right from the Start in their children’s schools.⏎Statistics courtesy of:⏎This is my first time visit at here and i am truly⏎impressed to read all at alone place.<|endoftext|>This section is dedicated to the debate regarding ke
-
th group. Rexroth is a manufacture of factory automation equipment; their site has a great section on roll-to-roll plastic electronics assembly technologies. It is fascinating to look through the process – it gets my imagination running on what the future of light
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 have a 35-foot, 24-inch, 36-inch, and 16-inch ceiling on my home. The middle middle, middle, and middle, and middle, and middle, and middle, and middle
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is different, so it’s important to recognize that what is appropriate and works for one child might not for another. But these are good natural strategies to implement and see if they can help your child.⏎1. A Clean Diet
-
perception and expression of information⏎Difficulty understanding/interpreting what one hears or reads⏎Difficulty remembering information⏎Difficulty seeing or processing/understanding what is seen⏎For additional information please contact the Counsel
-
not “exercise.” (This can have a VERY large impact on your daily caloric burn. It’s the difference between sitting on your butt on the computer or watching TV and moving, just doing something, anything.)⏎- TEF-thermic
-
or binoculars to see them. They move too fast for most telescopes to track anyway.⏎On the
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
- L21:2151 excites
- L18:2988 suppresses
- L21:2928 suppresses
- L21:2562 excites
- L19:943 excites
- L20:2725 excites
- L21:1592 suppresses
- L19:930 suppresses
- L21:2920 excites
- L20:1142 suppresses
- L10:277 weakly excites
- L21:2622 weakly excites
- L18:2008 weakly excites
- L13:2362 weakly suppresses
- L16:2727 weakly suppresses
- L20:1532 weakly excites
- L21:1394 weakly suppresses
- L19:525 weakly suppresses
- L19:2961 weakly excites
- L21:2319 weakly excites
- L19:3022 weakly suppresses
- L14:1973 weakly excites
- L9:1557 weakly suppresses 'the' after 'disease of' or 'infection of'
- L14:2898 weakly excites
Feeds later layers
- L24:1202 strongly excites
- L25:1085 excites
- L23:1827 excites
- L23:2098 excites
- L26:2829 excites
- L27:2831 excites
- L27:1431 excites
- L27:1133 excites
- L24:963 suppresses
- L23:738 excites
- L23:1482 excites
- L24:1379 excites
- L24:1606 excites
- L24:738 weakly excites
- L23:1454 weakly excites
- L26:966 weakly excites
- L23:721 weakly excites
- L23:622 weakly suppresses
- L24:2299 weakly excites
- L23:110 weakly suppresses
- L25:1032 weakly suppresses
- L27:1282 weakly excites
- L24:589 weakly excites
- L24:391 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
- 看
- see
- 听
- seeing
- 聽
- 看他
- sees
- see
- 阅读
- 看了
Suppresses
- nyder
- راح
- Rank
- 表达
- ость
- minster
- 逆转
- SetTitle
- paint
- bury