'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. 3 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 fires on the word "of" when it comes right after a noun and links it to a following noun, like "silhouettes of" or "a relationship of". It is quiet, or pushed the other way, on "of" in the phrase "get hold of". Steering did not give a clear theme.
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.
-
names and call them by name when participating in class. This helps to build a relationship of trust giving the student one more area of attachment to the class psychologically.⏎The way
-
those dealing in, acquiring, buying or borrowing items of cultural property have a duty to:⏎- keep abreast of legal issues;⏎-
-
K S M S; Balasubramaniam, V M; Niranjan, K; Knorr, D⏎Consumers increasingly demand convenience foods of
-
a polygon or topologically of a mesh. In the context of spatial data and Geographic Information Systems, nodes are usually referred to as points, non-closed connections of edges as lines, and meshes as polygons.⏎- More detailed information for vector data
-
Words in Another Language? There are as many ways to learn vocabulary as there are successful language learners throw this there are successful language learners Throw this out to a room of polyglots and watch as they they initiate a battle royale style fight to
-
on its histology by investigators. Extraovarian primary peritoneal carcinoma (EOPPC), serous surface papillary carcinoma, papillary serous carcinoma of the peritoneum, extra ovarian Mullerian adenocarcinoma,
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 feel like of the day of, and I'm not sure if I'm going to. It's been a bit of while, and I'm not sure if I'm going to. I'm not sure if I'm going to.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
only be a matter of time before we got hold of a dead animal - because they are hunted - and once we had and we started
-
energised and that the actual skin seems better.⏎If you have any inquiries pertaining to where and how to use Shapewear for women, you can get hold of us at our own web-site.⏎One technique to
-
or rejected by the college.⏎Get as much feedback as you can. Write for more than one class, and be sure that your classmates know how to get hold of you in the event you have any difficulties with your paper. Always ask for help
-
forget the painkillers in your handbag on the floor or on the bedside table!⏎The following video demonstrates how easy it is for a child to get hold of medication or household cleaning products and swallow them:⏎Take our quiz to find out if
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
- L0:2726 excites
- L0:897 weakly suppresses
- L1:725 weakly suppresses "to" in "referred to"
- L2:914 weakly excites
- L0:2158 weakly suppresses
- L2:895 weakly suppresses
- L2:2689 weakly suppresses
- L1:2733 weakly excites
- L0:1 weakly suppresses
- L2:2415 weakly suppresses
- L2:750 weakly suppresses
- L2:1454 weakly suppresses
- L0:1220 weakly suppresses
- L1:1473 weakly excites
- L0:1543 weakly suppresses
- L0:2952 weakly excites
- L2:2048 weakly excites
- L2:1259 weakly suppresses
- L1:455 weakly suppresses start-of-document end token
- L2:2397 weakly suppresses
- L0:1987 weakly suppresses
- L1:1122 weakly excites
- L1:491 weakly excites
- L0:560 weakly suppresses
Feeds later layers
- L16:1088 excites 'of' after a noun phrase, as in 'of course'
- L17:317 excites 'of' after a noun phrase
- L19:715 excites
- L9:2264 excites 'of' after superlatives and 'the X of'
- L5:363 excites
- L11:2207 excites "of" after a noun in trademark and title phrases
- L17:1835 excites
- L10:469 excites 'of' after a noun phrase
- L8:1350 excites
- L6:357 excites
- L15:785 excites 'of' in set phrases like 'of course'
- L15:2988 excites
- L22:229 excites "of" after a noun in "X of Y"
- L7:790 excites
- L8:1003 excites "the" or "of" before a noun phrase
- L14:2417 weakly excites
- L17:1507 weakly excites
- L4:4 weakly excites
- L8:349 weakly excites 'of' inside title-like noun phrases
- L4:3030 weakly suppresses
- L4:501 weakly excites "of" in "in terms of"
- L7:2705 weakly excites
- L17:1417 weakly excites
- L20:1972 weakly excites 'of' in 'of course'
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
- //!⏎
- empowerment
- Great
- awesome
- erta
- great
- course
- Great
- ós
- //!
Suppresses
- -driver
- ktör
- �
- bler
- 为了
- -divider
- isser
- eos
- pour
- �