"the" after abstract topic nouns
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 "the" in lists or phrases about broad subjects, such as "the importance of the" or "Fans of science, mathematics, the". When turned up, the model repeated the word "concept" over and over.
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.
-
philosophical figures: Archimedes, Sir Isaac Newton, Einstein, and more.⏎- Powers and principles in physics: Centrifugal and centripetal forces, the power of inertia, potential energy, kinetic energy, action and reaction law, buoy
-
, environments, fossils, and dinosaurs, parts of the earth, natural resources, how the earth moves, light and shadows, how things move, heat, and the systems of the body. Science process skills are developed by participating in hands-on activities
-
as such, the frequent references to it seem more like self-promotion than an attempt to contextualize and/or bolster his arguments. Fans of science, mathematics, the liberal arts, and the value of a well-rounded education may find themselves echoing Daniel
-
Century. It will focus on a number of themes, including the impacts of war on the U.S., civil rights and expanding notions of citizenship, the expansion of the role of the federal government, and the rise of the U.S. as a
-
this year’s edition of the UCD Connections magazine. In this article, Dr Tom Ryan writes that “87% of adults in Ireland recognise the importance of the environment
-
critically.⏎- Through bookmaking, students learn to express and extend their knowledge of concepts-either abstract concepts, such as friendship, or concrete concepts, such as the food chain.⏎Common Core Standards⏎This resource has been aligned to the Common Core
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: it is important to start with the use of a good, well-being being a key reference. It is important to include it with a good, well-being, and it is essential to do it with a good, well-being. So,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Water Buffalo (Bubalus bubalis),⏎- Guar (Bos gaurus), Rhino (Rhinoceros unicorns),⏎- Pigmy Hog (Sus salvanius).⏎Over 426 bird species are reported to be found in
-
NSW government that a huge. Jan 06, 2014 · Case study on: Ganga water pollution 1. Submitted by: Shradddha Samant MMS
-
of the land.⏎A quick search on the internet reveals that Lloyd George Street is in a popular part of the city with trendy shops, hotels, bars and The Lev Smadar, a highly regarded and comfortable arthouse cinema.⏎Unsurprisingly after
-
etc. There is no confusion in the first name. The middle name can be seen from example – Ram Prasad Srivastava. Here First Name is Ram, Middle name is Prasad and Last
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
- L21:614 excites
- L20:361 excites
- L17:30 suppresses
- L18:799 excites
- L9:24 excites "the" before a named organisation or title
- L19:427 suppresses
- L16:419 weakly excites
- L17:985 weakly excites
- L20:2795 weakly suppresses
- L18:666 weakly suppresses
- L18:602 weakly suppresses
- L16:792 weakly excites "the" inside a prepositional phrase
- L10:120 weakly suppresses
- L18:1041 weakly suppresses
- L14:169 weakly suppresses
- L10:1137 weakly excites
- L16:148 weakly excites
- L9:586 weakly excites
- L20:2826 weakly excites
- L16:775 weakly suppresses
- L19:2283 weakly suppresses
- L16:2062 weakly excites "other" and contrast words
- L14:2094 weakly suppresses
- L9:1009 weakly suppresses
Feeds later layers
- L26:613 weakly suppresses
- L23:2674 weakly excites
- L24:2470 weakly excites
- L27:1180 weakly excites
- L24:803 weakly excites
- L24:1811 weakly suppresses
- L23:176 weakly suppresses
- L26:532 weakly excites
- L25:2245 weakly excites
- L26:2930 weakly excites
- L23:937 weakly excites
- L23:670 weakly suppresses
- L25:508 weakly suppresses
- L24:405 weakly suppresses
- L24:2485 weakly excites
- L26:405 weakly excites
- L27:1755 weakly suppresses
- L27:528 weakly excites
- L27:1846 weakly excites
- L25:2206 weakly excites
- L26:2836 weakly suppresses
- L24:757 weakly suppresses
- L23:2785 weakly suppresses
- L24:2263 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
- concept
- 这位
- Double
- 这项
- 这条
- 这张
- 这个
- 这套
- 这么多
- Double
Suppresses
- aders
- 本报讯
- <IActionResult
- }->{
- insic
- adiens
- ört
- =row
- edium
- adians