"some of" before a list
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. 2 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 "of" right after "some", as in "some of the" or "some of these", where a list or group is about to follow. When it was turned up, the written text did not clearly change in a readable way.
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.
-
minerals, fiber, and antioxidants that can help to protect against a range of diseases, including cancer, heart disease, and diabetes. Here are some of the key health benefits of broccoli:⏎- May help to reduce the risk of cancer
-
children need specific occupational, physical, speech and behavior therapy. Some of these services are provided in the school district, but not all, not always,
-
and religion, perhaps no symbol is more ambiguous and double-edged than fire. Fire symbolism can simultaneously denote illumination and purification and destruction and pain. What are some of the main meanings of fire symbolism in literature and art over the centuries?⏎Fire-symbol
-
hierarchical animated tree visualization of trace links. Figure 1 below shows the user interface.⏎This capstone project has been a tremendous learning experience. The following are some of the things that were learned over the course of the 2 quarters
-
. Some of these plants are not good neighbors, and refuse to stay in our yards, with
-
. Some of the activities include making a jester hat, learning the chivalric code,
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 been. I have been. I have been. I have been. I have been. I have been. I have been. I have been. I have been. I have been. I have been. I have been.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
gum disease, and tooth loss. Here are some tips on how to take care of teeth.⏎1. Brush
-
to help improve the stability, function and health of an individual's teeth; otherwise, many people would be at higher risk for gum disease, tooth decay and tooth loss because of improper teeth positioning in their mouth,
-
covers the teeth and causes yellowing. When this plaque isn’t removed, gingivitis can advance to periodontitis, which can attack bone and cause tooth loss. The ideal candidate for gingivitis treatment with Dr. Boulard is
-
FC) have developed a very thin, flexible supercapacitor with high energy density.⏎The team from UCF's NanoScience Technology Center figured out a way to use two-dimensional materials as super
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
- L1:1623 excites
- L1:991 excites
- L1:992 weakly excites
- L1:2446 weakly excites
- L1:685 weakly excites
- L0:1017 weakly suppresses
- L1:2835 weakly excites
- L1:800 weakly suppresses
- L1:2340 weakly suppresses
- L1:2732 weakly excites
- L1:2431 weakly excites
- L0:1305 weakly suppresses
- L0:186 weakly excites
- L1:1583 weakly excites
- L1:2708 weakly excites
- L1:2372 weakly excites
- L1:146 weakly excites
- L1:2616 weakly suppresses
- L1:2447 weakly suppresses
- L0:1059 weakly suppresses
- L0:2831 weakly excites
- L1:330 weakly excites
- L1:1510 weakly suppresses
- L1:2392 weakly excites
Feeds later layers
- L7:885 excites
- L13:2296 weakly suppresses
- L5:1413 weakly suppresses
- L22:1275 weakly excites
- L11:2158 weakly suppresses
- L5:285 weakly excites
- L12:2666 weakly excites
- L18:327 weakly suppresses
- L4:148 weakly excites
- L16:2310 weakly excites
- L19:2439 weakly excites
- L18:364 weakly excites
- L3:177 weakly excites
- L20:2307 weakly suppresses "some" in "some kind" and "at some point"
- L3:124 weakly excites
- L25:1891 weakly excites
- L13:474 weakly excites
- L19:393 weakly excites
- L4:121 weakly excites
- L5:78 weakly suppresses
- L20:263 weakly excites
- L27:557 weakly excites
- L19:1538 weakly excites
- L19:2303 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
- zel
- ViewItem
- organic
- 部副部长
- ificant
- einz
- GetMessage
- 쩍
- 石
- atica
Suppresses
- ervations
- enarios
- üyor
- ersions
- elian
- arsing
- 苴
- points
- iyor
- erties