"for" introducing a purpose or source
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. 6 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 "for" when it links a thing to what it is for, sought from, or meant for, such as "looking to English sources for portraiture" or "turning to animals for food". Steering it up or down did not produce a clear change in the writing.
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.
-
as in his other works of the time, the artist employed formulaic poses, gestures, and dress, a result, in part, of looking to English sources for portraiture, but also a means of meeting the many requests for commissions.
-
. (See Speed Force Powers section.) However, the Speed Force is not the source from which other DC characters with super speed such as Superman or Captain Marvel derive their powers.⏎Writer John Byrne maintained modest abilities for the speedster character Danny Hilltop
-
do prey on other animals. And, now, as their traditional food sources are becoming more scarce because of human development, Dingoes are turning to domesticated animals for food, pitting themselves against the sheep farmers.⏎“Quite often they’ll go
-
0)⏎See Corbett (2000:173) for languages where mass nouns take plural rather⏎than singular morphology (and a few languages
-
vital so that you have a timing grid where you can follow the notes as they pass. Refer to the sheet ‘Reading Music: The Very Basics - Examples’ for some simple lines to get all this information working for you.⏎For now, when
-
nitrogen deficiency (yellowing leaves) and respond with quick side-dressings of a nitrogen-rich fertilizer such as fish emulsion. Contact your local County Extension office for controls of common corn pests such as corn earworms and European corn borers
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 fact that the community is the first to the sun, and the second to the sun. In the first row, I have 1, 2, 3, and 4, and the second to the sun. If the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Cochinos, these projects would not have been possible without Anthony Ives. A former San Francisco stock trader who at one point was managing 50 billion dollars for the state of California, Ives found his way to Honduras after Sept.
-
Labor Statistics.⏎The Census Bureau, which tracks annual wages, found women who worked full-time, year-round in 2010 made 77 cents for every dollar men earned across the country. This comparison includes all male and female workers
-
the days of rest, some slaves would hold quilting bees, with both men and women. It was also sometimes tradition that slaves could keep the money they earned for the sale of goods during the holiday.⏎While the holiday season was meant to be
-
Gray Otis House⏎There are three houses named the Harrison Gray Otis House in Boston, Massachusetts. All were built by noted American architect Charles Bulfinch for the same man, Harrison Gray Otis.-First Harrison Gray Otis House:
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
- L5:1090 excites
- L4:963 weakly excites
- L1:1578 weakly suppresses
- L3:381 weakly excites 'for' in research institute names
- L5:3026 weakly excites
- L5:1308 weakly suppresses
- L4:1709 weakly suppresses
- L5:50 weakly excites
- L3:2843 weakly excites
- L3:1631 weakly suppresses
- L1:789 weakly suppresses
- L2:75 weakly suppresses
- L1:2479 weakly suppresses
- L3:1711 weakly suppresses comma inside large numbers
- L3:2567 weakly excites
- L4:1222 weakly suppresses
- L5:2022 weakly suppresses
- L4:2718 weakly excites
- L2:2820 weakly excites
- L5:2841 weakly excites
- L1:2008 weakly suppresses
- L4:2265 weakly suppresses
- L1:1860 weakly suppresses
- L0:2521 weakly suppresses
Feeds later layers
- L15:1386 suppresses prepositions after passive verbs
- L9:654 suppresses
- L14:1547 excites
- L14:1192 suppresses
- L9:2390 suppresses
- L8:2715 excites
- L12:2276 excites
- L12:2756 excites "of" after a noun of quantity or container
- L10:2278 suppresses
- L9:1571 weakly excites
- L12:1916 weakly excites
- L9:2267 weakly suppresses 'with' after exchange nouns
- L15:2207 weakly suppresses
- L7:428 weakly excites
- L17:2197 weakly suppresses
- L7:1579 weakly suppresses
- L8:447 weakly excites
- L16:1877 weakly excites
- L7:14 weakly suppresses
- L15:2225 weakly excites
- L11:361 weakly suppresses
- L12:1642 weakly excites
- L9:1562 weakly suppresses 'by' before a byline or agent
- L19:544 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
- 更多信息
- 筏
- EventArgs
- information
- salvation
- 机电
- 入党
- 的信息
- 收益
- 解答
Suppresses
- tsy
- strar
- ь
- embro
- theast
- ント
- umpy
- ROUP
- actory
- aha