abstract head nouns after 'the' or 'some'
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. 12 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 abstract nouns that complete a phrase, such as 'law' in 'supreme law', 'faith' in 'Catholic faith', or 'knowledge' after 'some'. It is most negative on words like 'themselves' and 'one another'. Turning it up did not give a clear, matching change in the 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.
-
it is 180°, and for a class C amplifier it is less than 180°.<|endoftext|>The Constitution of Nigeria is the supreme law of Nigeria. It sets out the framework for government and all its activities, including
-
essential qualities and characteristics which are transmitted over time.I am reminded of GK Chesterton's notion of The Thing. The idea that there is some thing called the Catholic faith and it makes sense of every other thing. There is not one area of life
-
Gertrude has some knowledge of what happened, whether she is mentally unbalanced and is in denial, or
-
improve your students' grades. In the Course Activity Related to Grades report, also known as learning analytics, you can view how active your students are in your course materials. This activity has been shown to correlate strongly with course outcomes. Students who are
-
most prominent unique personal characteristics of the G&T are: being Critical, Persistent & Independent of thought & judgement. Studies conclude that there are at least 18 areas of sensitivity. (CHART by Dunn & Price)⏎Also see posts: “
-
's Medicare and Medicaid programs are willing to cover. Lichtenberg writes that there are two ways to improve health-care quality: by ensuring that patients get "best practice" care and by "shifting the technological frontier" through the use of new
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: "I think the problem with the problem of the problem of the problem of the problem of the problem. The problem with the problem. The problem of the problem. The problem with the problem. The problem of the problem. The problem.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the church? This question, frequently discussed in THE EXAMINER, is important because it addresses the basic issues of Christlike relationships, Christlike respect for one another, Christlike service of the powerful to the powerless, and the individual responsibility
-
was born.⏎Cloning is not new to nature, just new to humanity. Many creatures practice asexual reproduction, or parthenogenesis, producing exact copies of themselves. Many plants reproduce themselves by sending out roots
-
in thinking it made the speaker sound self-righteous.⏎Mark, here are your answers.⏎Illeism Is the Habit of Referring to Yourself in the Third Person⏎This verbal tic is known as "illeism." That's the habit of
-
we want to retrieve the correct value from the data table above.⏎Have you ever built an app based on Excel?⏎Method 1: The easiest way (VLOOKUP & MATCH)⏎Probably the easiest – or at least the shortest – way of
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
- L10:850 suppresses
- L10:2679 suppresses
- L11:305 suppresses
- L10:164 excites
- L10:483 excites
- L11:1482 excites
- L11:2545 excites
- L9:64 excites
- L2:558 suppresses
- L11:420 excites
- L10:1390 weakly suppresses
- L10:324 weakly suppresses
- L10:2645 weakly excites
- L7:1402 weakly excites
- L9:868 weakly suppresses
- L10:1103 weakly suppresses
- L11:865 weakly excites
- L2:314 weakly suppresses
- L9:383 weakly suppresses
- L11:1542 weakly suppresses
- L10:2675 weakly excites
- L10:2431 weakly excites
- L6:2652 weakly excites
- L9:63 weakly excites
Feeds later layers
- L15:837 strongly excites
- L13:1933 suppresses
- L14:1651 excites
- L14:1364 suppresses
- L13:2783 excites
- L14:2126 excites
- L13:1667 excites
- L13:1807 suppresses
- L14:525 suppresses
- L16:335 suppresses
- L13:971 excites
- L13:1470 excites
- L14:1109 excites
- L16:1144 excites
- L13:2133 excites
- L13:167 suppresses
- L15:878 suppresses
- L13:1843 excites
- L13:193 suppresses
- L13:1073 excites
- L14:868 excites
- L13:2056 suppresses
- L14:576 excites
- L21:2746 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
- trì
- newsp
- jective
- idata
- 面前
- Quốc
- mers
- cı
- »⏎⏎
- 这支
Suppresses
- 矣
- uds
- いません
- ubishi
- Raised
- ),$
- auled
- raised
- Raised
- elt