comma after an opening clause
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. 18 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 comma that ends an introductory phrase such as 'When he arrived,' or 'If I ask for something,' just before the main sentence starts. Turned up, the model fell into repeated phrases like 'I feel, I feel', which does not clearly relate to this trigger.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
found on beaches worldwide. When one thinks of the word recycling, it is newspapers, cardboard, glass, steel, aluminum foil, aluminum cans,
-
stick to it.⏎4. Start thinking more positively in order to build self-confidence.⏎How to identify and challenge negative thoughts:⏎When we’re faced with negative thoughts, it can be hard to know where to start. We may find it difficult to
-
genuine a blessing that phrase is. Bed bugs can make your life miserable and can be very difficult to get rid of.⏎When most of us think of bed bugs, we think of one species, but did you know there are more than ninety?
-
years were over he had said all that he had to say, for his conversation was limited, and he determined to return to his own castle. When he arrived, he saw the children playing in the
-
Jesus and his disciples on many occasions. Surely, Jesus would fulfill their request.⏎But that isn’t what happened.⏎When Jesus received word about Lazarus’ condition, he stayed where he was for the next two days. It wasn’t until the
-
code "reads" easily, and yet it has the flexibility to contain all the different things I need to return. If I ask for something that isn't there, I get an exception, which is life as usual for a Python programmer.⏎if
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: and then, and then, and then, and then, and then, and then, and then, and then, and then, and then, and then, and then, and then, and then, and then, and then,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
any reconfiguration event, there is no need to run this command interactively.⏎For more information, see the following references:<|endoftext|>The United Kingdom, with its constituent parts, including
-
up on the Web—until I saw one geek-speak phrase on the home page, so small I almost missed it: API Documentation. To understand its significance, let me tell you how I got subway schedules on my iPhone.⏎Just a few
-
link a person’s thoughts, conceptualizations, and memories with a computer. An Imagination Interface, perhaps. By any name, it doesn’t take much time to realize that you have always wanted one
-
one time link to download ebooks to your computer. Orders paid by PayPal require up to 8 business hours to verify payment and release electronic media. For immediate downloads, payment with credit card is required.⏎This math unit presents four activity pages in which
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
- L13:1232 excites
- L17:1426 excites
- L16:1886 excites
- L4:2769 suppresses
- L17:1845 excites
- L11:1084 excites
- L8:410 suppresses
- L9:1579 suppresses
- L13:530 suppresses commas after a 'since' clause
- L14:1209 suppresses
- L15:2295 suppresses
- L13:706 excites
- L17:3021 excites
- L6:1485 excites
- L15:1830 suppresses
- L6:2233 excites
- L17:2397 excites
- L13:1273 suppresses
- L16:2210 suppresses
- L13:921 weakly excites
- L6:2843 weakly suppresses
- L14:2927 weakly suppresses comma after a conditional lead-in clause
- L17:2960 weakly suppresses
- L16:968 weakly excites
Feeds later layers
- L22:1551 weakly excites
- L20:2891 weakly excites
- L21:1691 weakly excites "at" in "at first" and "initial" words
- L27:910 weakly excites
- L20:2484 weakly suppresses
- L22:689 weakly excites comma closing an intro clause
- L24:33 weakly suppresses
- L19:2726 weakly suppresses
- L20:501 weakly suppresses
- L24:13 weakly suppresses
- L19:2508 weakly excites
- L21:2941 weakly suppresses
- L22:2314 weakly excites
- L19:2440 weakly suppresses
- L20:1281 weakly suppresses
- L21:1044 weakly excites
- L23:1413 weakly excites
- L20:610 weakly suppresses
- L25:299 weakly suppresses
- L19:2023 weakly suppresses
- L21:1612 weakly excites
- L21:2901 weakly suppresses
- L22:1391 weakly excites
- L24:945 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
- 首先
- 首先要
- INGS
- 最先
- zunächst
- 务必
- liğinde
- usting
- aturas
- 必定
Suppresses
- その後
- [str
- _refer
- ضاف
- InSection
- そして
- onnen
- ��
- luyện
- roduce