'on' introducing a location or surface
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. 8 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 lights up on the word "on" when it says where something is, such as "printed on elastic fibers," "posted on the website" or "on the trail." Its negative side fires on "wall" in phrases like "stomach wall." Turning it up gave outputs that did not clearly relate to location.
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.
-
credit score, which the likes of ClearScore, Equifax and Call Credit allow you to do for free, and taking a few easy steps such as getting on the electoral register, you can
-
accessible to employees for all hazardous chemicals in their workplace. This may be done in many ways. For example, employers may keep the SDSs in a binder or on computers as long as the employees have immediate access to the information without leaving their work
-
our place in the solar system.⏎It took those roving teams months to assemble at designated points at the Arctic tip of Norway, in Canada's Hudson Bay and on a small island in the South Pacific. They battled the elements, disease, hunger
-
Control your speed⏎- Warn others when passing⏎- Keep dogs on a leash at all times for their protection, and for yours⏎- Respect all wildlife encountered on the trail<|endoftext|>In 1847, when Mormon pioneers arrived in the
-
opposites."⏎Opposites are everywhere in our world. A day at the park will reveal busy kids and active animals doing opposite actions. There, on the slide: Sam goes up as Hiroko zips down. And there,
-
school district has a list of bus rules and regulations posted on the district website to assist parents and staff in making sure students are safe. OSD
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 thinking about how I can use the power of the numbers to make the difference between the first and second and third digits, and the power of the numbers to make the difference between the first and second and third digits. So,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and break food down into smaller particles using chemical and mechanical digestion.⏎The stomach can perform these roles due to the layers of the stomach wall. These are the gastric mucosa, submucosa, muscularis extern
-
stomach to distend upon the entry of the food. A bolus of food enters the stomach from the esophagus. The various tissue layers of the stomach wall then combine their functions to digest the bolus into a viscous, pulpy
-
you still find it daunting, why not brush up on your histology basics first?⏎- Stomach wall⏎- Muscularis externa⏎The stomach wall consists of 4 layers of tissue. From deep (external) to superficial (
-
War. Green safely ejected and was rescued from the Yellow Sea by an Air Rescue Service SA-16 Albatross flying boat.⏎Captain Fedorets closed on McConnell from
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
- L4:1016 excites
- L4:3030 excites
- L7:2824 excites
- L1:1194 suppresses
- L5:3024 excites
- L4:1406 weakly excites
- L2:2307 weakly excites
- L7:1504 weakly suppresses
- L6:2392 weakly suppresses
- L5:3065 weakly excites
- L7:1383 weakly excites
- L5:1431 weakly excites
- L7:2655 weakly excites
- L7:1551 weakly suppresses
- L2:1226 weakly suppresses 'in' after belief verbs
- L4:2079 weakly suppresses
- L2:1798 weakly suppresses
- L7:2137 weakly excites
- L7:1780 weakly suppresses
- L7:2758 weakly suppresses
- L4:2109 weakly excites
- L7:1877 weakly excites
- L5:1644 weakly suppresses
- L1:2721 weakly suppresses
Feeds later layers
- L14:1122 strongly excites
- L18:749 excites
- L22:2554 excites
- L20:1905 excites
- L11:1187 excites
- L25:2569 excites
- L13:2135 excites
- L9:3031 suppresses
- L17:2101 excites
- L10:3068 excites
- L16:2852 excites 'in' after a noun or participle
- L12:2638 weakly excites
- L19:1345 weakly excites
- L12:2289 weakly excites 'of' after part-whole nouns
- L12:2833 weakly excites
- L14:1728 weakly excites "of" after a numbered section or title
- L17:2877 weakly excites
- L11:1369 weakly excites
- L19:252 weakly excites
- L18:2836 weakly suppresses
- L9:1025 weakly excites
- L19:649 weakly excites
- L12:2889 weakly excites
- L14:427 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
- ensively
- platforms
- isode
- 战场
- shelves
- naments
- реб
- ду
- ross
- riday
Suppresses
- pé
- 誉
- administr
- onian
- ế
- ónica
- onica
- xcf
- onyms
- amb