'of' after a noun phrase
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. 10 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 'of' when it comes right after a noun, as in 'the time of Moses' or 'the 34th day of 2013'. The same word in some other 'of' phrases pushes it strongly negative, so it is not a clean signal. Turning it up produced rambling text that did not clearly relate to the word 'of'.
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.
-
hearts or arteries preserved enough to study. Of those, nine had evidence of blockage from atherosclerosis. “This disease has been around since before the time of Moses,” said Randall Thompson
-
of Pakistan borders Afghanistan, India, Iran. Gross domestic product (GDP) is of about 315000 millions of dollars dollars.⏎K2 is
-
.noaanews.noaa.gov⏎– Erich Luening⏎Print this page<|endoftext|>Today is Sunday, Feb. 3, the 34th day of 2013. There are 331 days left in the
-
s) have not been enumerated in any of the Census since 1951 till day”. The submission was in response to a plea by the State Government of Maharashtra seeking a direction to the Census Department to collect information on Backward Class of
-
After Einstein’s death, his brain was preserved for future study. Scientists were naturally curious to see how the brain of this genius compared with the brain of a person of ordinary intelligence. Would there be an abundance of neurons (grey matter) or some
-
creating "cricket reactors"—enclosed cricket breeding contraptions fed by recycled organic wastes like lawn clippings and kitchen scraps.⏎His apartment contains one of a few working prototypes. (His wife almost kicked him out when a few escaped,
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: the mind and the brain. The brain. The brain. The brain as a subject. The brain, the brain as a subject, the brain. The brain. The brain as a subject. The brain. The brain. The brain.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
) discusses the long term use of fossil fuels, further exploring the notion that certain reserves of oil, gas and coal should not be extracted and used due to concerns about rising
-
failure, the toxic components of lily of the valley affect your dog's heart. These components are called cardiac
-
lives. If women united internationally and with the wider struggles of working<|endoftext|>|Part of a series on|⏎|Types of creationism|⏎|
-
situated at the valley of the mountain. These statues represent the style of Phnom Da. Among the five
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
- L4:4 suppresses
- L9:1428 excites
- L8:1003 suppresses "the" or "of" before a noun phrase
- L4:1686 excites
- L5:363 excites
- L3:315 excites 'of' after a noun phrase
- L1:679 suppresses
- L2:895 weakly suppresses
- L6:357 weakly suppresses
- L9:2264 weakly excites 'of' after superlatives and 'the X of'
- L7:1152 weakly suppresses
- L4:315 weakly suppresses
- L8:668 weakly suppresses
- L5:384 weakly suppresses
- L3:518 weakly excites
- L2:2786 weakly excites
- L9:1446 weakly suppresses
- L5:1614 weakly suppresses
- L5:1404 weakly suppresses
- L1:2033 weakly suppresses
- L9:1942 weakly suppresses
- L8:688 weakly excites
- L6:915 weakly suppresses
- L4:561 weakly excites
Feeds later layers
- L22:229 suppresses "of" after a noun in "X of Y"
- L13:132 suppresses "of" after a noun
- L12:2757 suppresses "of" after a measure like inches or teaspoon
- L13:1062 suppresses
- L12:2022 suppresses 'of' after type or class nouns
- L15:801 suppresses 'of' inside organisation names
- L12:850 excites
- L15:232 excites
- L17:1835 suppresses
- L17:385 suppresses
- L14:2075 excites
- L16:805 suppresses
- L17:1507 suppresses
- L17:317 suppresses 'of' after a noun phrase
- L21:399 suppresses
- L14:298 suppresses
- L15:1554 weakly suppresses
- L20:283 weakly suppresses
- L19:207 weakly suppresses
- L17:2302 weakly suppresses 'in' after a noun
- L13:1876 weakly suppresses 'in' or 'on' before a place or date
- L15:395 weakly suppresses
- L17:1417 weakly suppresses
- L14:1519 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
- imer
- aux
- ifique
- 均有
- avan
- 断
- imientos
- 总有
- 总是
- 均衡
Suppresses
- sorts
- building
- CLUD
- ark
- .tie
- wards
- circuits
- closed
- @@⏎
- civil