'of' after amount or level 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. 5 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' after nouns of quantity or degree, like 'amount of', 'levels of' or 'numbers of'. The signal is fairly weak, and turning it up or down did not clearly change the written text.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
Celsius, and preferably 1.5 degrees Celsius, from pre-industrial levels. The aim is to use the best available scientific data to reduce the global amount of greenhouse gases, mainly carbon dioxide and methane emissions, in the Earth’s atmosphere.⏎
-
utable to this, the forest officials can not only restrict tribal and forest dwellers from accessing forest lands and produce but can also punish the whole community for the act of an individual.⏎Into the bargain, the government has gone a step further and shifted
-
, all the social and medical developments we’ve achieve thus far haven’t tackled the TB problem in its roots. Scandinavian countries have been able to dramatically decrease the numbers of people getting ill with TB long before antibiotic treatment existed, thanks to improvements in housing
-
bi-products such as saw dust and planer shavings. This production demanded large amounts of water, and thus the pump tower and a water line was established. Both pump
-
31 products tested in the study, at least three-quarters of waterproof mascaras, two-thirds of foundations and over half of eye and lip products had high amounts of fluorine. All 29 products that were tested for PFAS after showing
-
These lipids consist of ceramides, fatty acids and cholesterol that help to hold in water and prevent dryness. People with dry skin often have lower levels of ceramides. This affects the skin’s ability to retain moisture. When the
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: a long way, and then a 1. So, the website will be built from that. The problem is that the website is going to be built from that, and then the next part is going to be built from that. The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
1,000 striking plantation workers.⏎In 1959, rock-and-roll stars Buddy Holly, Ritchie Valens and J.P. "The Big Bopper" Richardson died in a small plane crash near Clear Lake,
-
the digital humanities. Through careful explanations of the field, examinations of terminology and themes, and illustrated case studies of diverse texts—from the Cyrus cylinder to the Eagles' "Hotel California"—Elaine Treharne and Claude Willan offer a clear
-
the Art and the Future Glossary References Author index Subject index About the Author.⏎(source: Nielsen Book Data)⏎'The fourth edition of Uwe Flick's "Introduction to Qualitative Research" remains the most comprehensive and thorough text in qualitative research
-
To access this title, visit your library in the app or on the desktop website.⏎Chronicles of the imagination Geographical⏎- Walta Evans "School Librarian"<|endoftext|>Wing Tsun or Wing Chun?⏎or Wing Chung
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
- L1:606 weakly suppresses
- L3:2860 weakly excites
- L4:1081 weakly suppresses
- L0:32 weakly suppresses
- L2:2057 weakly suppresses
- L4:2199 weakly suppresses
- L4:781 weakly suppresses
- L4:982 weakly excites
- L0:1965 weakly excites
- L3:1372 weakly suppresses
- L4:1998 weakly suppresses
- L4:1373 weakly excites
- L4:207 weakly suppresses
- L4:2415 weakly excites
- L4:2230 weakly suppresses
- L4:1835 weakly suppresses
- L4:2565 weakly suppresses
- L4:2190 weakly excites
- L1:1069 weakly excites
- L3:601 weakly suppresses
- L1:1222 weakly suppresses
- L2:1355 weakly excites "about" in "what about" questions
- L4:635 weakly suppresses
- L3:2540 weakly excites
Feeds later layers
- L23:2092 weakly excites
- L6:2259 weakly excites
- L24:1722 weakly excites
- L22:2743 weakly excites
- L6:2909 weakly suppresses
- L15:1354 weakly excites
- L7:601 weakly suppresses
- L20:2306 weakly excites
- L9:1934 weakly suppresses
- L24:1363 weakly excites
- L7:921 weakly excites
- L17:64 weakly suppresses
- L18:960 weakly excites
- L12:2245 weakly excites
- L18:1429 weakly suppresses
- L10:472 weakly suppresses
- L7:1684 weakly excites
- L14:555 weakly suppresses
- L20:2461 weakly excites
- L15:2627 weakly excites
- L23:2020 weakly excites
- L10:2586 weakly suppresses
- L25:1734 weakly excites
- L16:892 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
- ировки
- registry
- ак
- бу
- -outs
- ContentValues
- elling
- 승
- rus
- Smarty
Suppresses
- endet
- toughness
- ち
- aser
- �试
- itter
- offsetof
- primera
- ôte
- 提前