'one or more' before a count 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. 21 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 'or' in the phrase 'one or more', and on similar phrases like 'more than one'. When it was turned up, the model started repeating 'than' and 'multiple', so it seems tied to the idea of quantity.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
me that you specify where each of your posts falls amongst the categories listed below.⏎The same post can count for more than one category, but if you keep re-using the same handful of posts
-
good solution. For instance, here is a piece of a docstring that I wrote yesterday:⏎Return a dict (which I often call "response") with one or more of theThe
-
myself 50 Bitcoins to my address A" (plus transaction fees from the block).⏎- When you want to spend Bitcoins, you have to point to one or more transactions that show where you got them from, and say how many Bitcoins you
-
the shoulders of sows kept in intensive production systems, in particular in farrowing crates.⏎- Lameness: When lame, a pig is unable to use one or more limbs in a normal manner. Lame pigs often experience pain, and may
-
become mandatory when they have been incorporated into a business contract or incorporated into regulations.⏎What is a code ?⏎A code is a standard that has been adopted by one or more governmental bodies and has the force of law.⏎Why are standards effective ?⏎Stand
-
site matches the name of the publisher.⏎Omitting the city of publication. Include the the city of publication only if the version of the source differs when published in a different country. (Example: British editions of books
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 two-silloy (3,2,1,2,4,6,8,12,6,4,2,4,2,4,6,12,6,4,2,4,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a longtime favorite for sleep-deprived people around the world. Arguably, the most famous and relevant of these compounds is 1,3,7-trimethylxanthine, better known as caffeine.⏎It’s pretty common knowledge that
-
grasslands, and dry deserts.⏎Pads, flowers, and fruits⏎Flavonoids, octadecadienoic acid, and hexadecanoic acid⏎Most species of prickly pears have similar effective and safe
-
ROMATIC COMPOUNDS ARE THOSE ORGANIC COMPOUNDS WHICH HAVE BENZENE RING (CYCLOHEXA-1,3,5-TRIENE). AROMATIC COMPOUNDS ARE ALWAYS
-
components and patterns on both sides of the boards. A double layer PCB is used on devices such as vending machines, phone systems, car dashboards, etc.⏎- Multi-Layer PCB – These types of PCB have more than three layers 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
- L20:1724 strongly suppresses
- L16:2111 strongly excites
- L17:2309 suppresses
- L18:1732 excites
- L18:2927 suppresses
- L11:1291 suppresses
- L14:2783 suppresses the word 'same' in 'the same'
- L13:1507 suppresses
- L20:2636 excites
- L13:1090 suppresses
- L14:2018 excites
- L17:2556 suppresses
- L16:1640 suppresses
- L17:1786 excites
- L13:2199 excites
- L14:1990 excites
- L10:2300 suppresses
- L20:1788 suppresses
- L12:2116 suppresses
- L17:1727 excites
- L19:2929 suppresses
- L17:3065 excites
- L17:904 excites
- L17:1789 suppresses
Feeds later layers
- L26:1784 strongly excites
- L23:888 strongly excites
- L25:1183 strongly excites
- L27:810 excites
- L26:1692 excites
- L22:825 excites 'for' before a span of time
- L23:751 excites
- L25:1756 excites
- L24:1576 excites
- L23:538 excites
- L24:2350 excites
- L27:712 weakly excites
- L23:231 weakly excites
- L27:948 weakly excites
- L23:408 weakly excites
- L23:1543 weakly excites 'up' in 'up to' limits
- L22:816 weakly suppresses
- L26:1724 weakly excites
- L24:1690 weakly excites
- L24:680 weakly excites
- L25:472 weakly excites
- L22:1016 weakly suppresses
- L25:2409 weakly excites
- L26:1089 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
- multiple
- multiple
- Multiple
- Multiple
- 多个
- _multiple
- 多
- MULT
- 多家
- mú
Suppresses
- tram
- war
- 嗡
- ierge
- endencies
- onz
- 三
- 六
- 四
- ---------- ⏎