unclear: 'of' after a quantity noun
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. 8 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 mostly fires on the word 'of' after nouns that describe an amount or group, such as 'a ball of roots' or 'a set of', but the examples are mixed and some are unrelated. Turning it up gave unrelated text, so the steering result adds little.
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.
-
resources to learn more about Caddo history and culture, https://scholarworks.sfasu.edu/cgi/viewcontent.cgi?article=4012&context=ita⏎Visit the THC’s Learning Resource page for home-related lesson plans
-
Tree?⏎When pruning the roots of your bonsai, the first step is to remove it from its pot and use a root hook to disentangle its ball of roots.⏎Then, remove as much of the soil as you can with your fingers
-
risk for sepsis.⏎They surveyed stool samples from 14 patients with long ICU stays, including six people who stayed more than a month. Using next-generation sequencing of the microbes’ ribosomal RNA 16S subunit, the
-
|⏎|Listener||Contains a method that handles an event|⏎|Entity||Data object with identity (ID), for example “User”|⏎|Collection||Set of data objects|⏎|Value Object||Data object without identity, for example “Date
-
were gone."⏎Join to answer this question⏎Join a community of thousands of dedicated teachers and students.Join eNotes<|endoftext|>Is there something else you would
-
Oregon town of Hammond, near the mouth of the Columbia. He’s taking a boatful of customers into shallow water to hook sturgeon.⏎“If you drop off the ledge just
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: questions. I have been thinking about how to questions to get a job. I have been thinking about a question to get a job, and I have found a job. I am a life partner that I have a living job. I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a time, it seemed like myrtle spurge was a solid choice for “water-wise” gardens. The plant grows well in shade, requires little in the way of hydration, grows in disturbed environments, and looks attractive. But myrtle sp
-
: "We watched trucks unload sacks of flour into crowds that hadn't eaten in three days. I can still see Schim [sic], a wisp of a man, standing on the jeep silhouetted against the darkling sky quietly
-
a survivorship program designed to educate children and their families about the challenges they may face as childhood cancer survivors -- has awarded $125,000 in college scholarships to 38 cancer survivors.⏎To learn more about the<|endoftext|>The
-
, questions and answers of programming and scripting languages:<|endoftext|>RALEIGH — New research has found that North Carolina’s shellfish industry provides over $27 million in economic impact and 532 jobs in the state. Until 20
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
- L6:2862 suppresses
- L7:1208 suppresses
- L6:2366 excites
- L6:2593 suppresses
- L6:1407 suppresses
- L6:398 excites
- L7:2164 suppresses period inside country abbreviations
- L6:1789 suppresses
- L6:1410 excites
- L6:2873 weakly suppresses
- L2:686 weakly excites
- L6:458 weakly suppresses
- L7:2414 weakly excites
- L1:847 weakly excites
- L6:1041 weakly excites
- L5:1403 weakly excites
- L4:2085 weakly excites
- L6:2635 weakly excites
- L4:1567 weakly suppresses
- L6:2224 weakly excites
- L6:1767 weakly suppresses
- L5:1982 weakly excites
- L6:2785 weakly excites
- L6:1892 weakly excites
Feeds later layers
- L10:167 suppresses
- L14:1240 suppresses
- L19:663 suppresses
- L14:1913 suppresses
- L17:985 suppresses
- L12:1561 suppresses
- L17:1241 suppresses
- L18:77 suppresses
- L12:1916 suppresses
- L20:2519 suppresses "of" after quantity nouns (millions of, amount of)
- L13:1659 weakly suppresses
- L14:2510 weakly suppresses
- L14:312 weakly excites
- L19:2711 weakly suppresses
- L13:1370 weakly suppresses
- L13:330 weakly excites
- L15:537 weakly suppresses middle digits of numbers
- L16:1122 weakly suppresses
- L15:1386 weakly excites prepositions after passive verbs
- L11:810 weakly excites
- L13:975 weakly suppresses
- L17:199 weakly excites
- L12:2680 weakly excites
- L20:337 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
- Sheet
- INET
- 这位
- 性强
- ATORY
- 族自治州
- eneg
- aire
- IGATION
- istine
Suppresses
- wdx
- òng
- fuels
- быт
- ymi
- 缢
- ewitness
- 年来
- cura
- 篑