'the' after a preposition or verb
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' when it follows a preposition or action word, such as 'through the', 'for the', or 'reap the'. It is weaker and less clear on 'the' right after 'is'. The steering outputs do not show an obvious theme.
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.
-
work in progress. We shall supplement it as we use the Actoz°° system created and developed by Prof. Jean-Pierre Dedieu http://en.wikipedia.org/wiki/Jean-Pierre_Dedieu, senior research fellow at CNRS http
-
more effectively engage with, support, and meet the needs of youth and young adults of color whom they serve. The brief also presents
-
to America's melancholia.⏎The United States in early 1964 was a society on the brink of profound transformation.⏎Rock-and-roll music, through the auspices of Elvis Presley, Little Richard, Chuck Berry and others, had
-
program designed to help all families - whether they be single or married parents, grandparents, aunts or uncles, foster or adoptive parents or anyone responsible for the care of children aged 10 - 15 years old. Families in
-
emphasis is placed on a narrow set of academic tasks, students fail to develop life skills. MIT graduates will first need to lead balanced personal lives before they can reap the rewards of professional success.⏎To develop solutions that promote work/life balance, we
-
th century Christian Europe willingly looked upon him as an international judge.⏎He was renowned for his charity. “The peace and blessings of the realm come to us through the poor,” he would say. Beggars were fed from his table; he
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: this is a 17th- and 18th-century English writer, so can we have a section of this writer's world in a modern, English, and thus, a now-both-both-both-both-
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’m a fan of physicist Richard Feynman, and one of his most memorable quotes explains that “Science is the belief in the ignorance of the experts.”⏎Indeed, the “experts” of
-
trace, or are glued from coffee filters or produced using a 3D printer at home: Necessity is the mother of invention, but can sometimes be considered a crime. With the reprocessing
-
www.healthscout.com/template.asp?id=669525<|endoftext|>It is certain, in any case, that ignorance, allied with power, is the most ferocious enemy justice can have.⏎– James Baldwin⏎Four decades of neoliberal
-
2014.<|endoftext|>The more I read about this, the more confused I become. Illuminance, as I understand it, measured in lux, is the perceived brightness of light on a surface. It
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
- L17:30 suppresses
- L20:361 excites
- L16:1851 suppresses
- L19:427 suppresses
- L16:792 excites "the" inside a prepositional phrase
- L21:614 weakly excites
- L20:219 weakly excites "the" after time or place prepositions
- L16:419 weakly excites
- L2:2526 weakly suppresses
- L21:20 weakly suppresses
- L17:1346 weakly excites
- L9:240 weakly excites
- L18:602 weakly suppresses
- L4:2324 weakly excites
- L5:2324 weakly excites
- L1:2023 weakly suppresses
- L19:1175 weakly excites
- L18:799 weakly excites
- L13:1975 weakly excites
- L19:1818 weakly excites
- L20:2723 weakly suppresses
- L6:443 weakly suppresses
- L17:1771 weakly excites
- L1:1442 weakly excites
Feeds later layers
- L27:480 weakly excites
- L27:1354 weakly excites
- L27:730 weakly suppresses
- L24:1439 weakly suppresses
- L26:430 weakly excites name fragments inside split words
- L26:374 weakly suppresses
- L25:1814 weakly suppresses
- L26:1046 weakly suppresses
- L27:2150 weakly excites
- L23:1557 weakly excites
- L23:1496 weakly excites
- L26:2134 weakly suppresses
- L26:2593 weakly suppresses
- L24:319 weakly suppresses
- L27:1162 weakly excites
- L24:641 weakly suppresses
- L27:2336 weakly excites
- L26:196 weakly suppresses
- L23:2501 weakly suppresses
- L25:1874 weakly excites
- L26:1542 weakly excites
- L27:1670 weakly suppresses
- L23:2952 weakly excites
- L27:1750 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
- unique
- >\
- needs
- usive
- >-
- uess
- ",-
- >null
- lush
- (unique
Suppresses
- yms
- effect
- 做什么
- 媾
- يون
- Become
- [src
- Aspect
- jabi
- ction