comma after 'sometimes' or 'the truth'
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. 16 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 a comma that comes right after a short opening phrase like 'Sometimes,' or 'The truth,' or 'Not only does it work,', where a contrast such as 'but' or 'however' is likely to follow. Turning it up gave repetitive filler, and turning it down gave rambling, so the steering effect is not very informative.
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.
-
chicken and eggs are a good bet.⏎- Not all carbs are created equal. Consume complex carbs, like whole grains. Avoid sugar and white flour. Sorry,
-
difference.⏎Quality assurance is process oriented. It is all about preventing defects by ensuring the processes used to manage and create deliverables works. Not only does it work, but is consistently followed by the team. Moreover, QA is about engineering processes that
-
you wish to move up, you’ll need to prove again and again that you know how to navigate technology. Sometimes, having that little bit of extra knowledge can really make a difference in your salary and
-
to climate change in the future? We explore the invention of cooking meat. You’ll find out why the advancements of cooked food helped our brains grow. All that,
-
ensure they are not cancerous.⏎Sometimes, the doctor will suggest the individual wait to let the polyps go away on their
-
ing, canoeing and kayaking?⏎Is there, actually? Is rowing, canoeing and kayaking very different from each other? The answer is yes, they are. These are not three different techniques of operating the same boat. The
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 am:1:1000000000000000000000000000000000000010000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
been easier!⏎Note: CD-ROM/DVD and other supplementary materials are not included as part of eBook file.<|endoftext|>is a former municipality in Akershus⏎, which lends its name to a municipality and a county in Norway⏎. The name
-
,000 square kilometers, are the only source of protection for<|endoftext|>1911 Encyclopædia Britannica/Diamond⏎DIAMOND, a mineral universally recognized as chief among precious stones; it is the hardest, the
-
Cell Biology⏎Chlamydomonas; Dynein ATPase; Flagella; Microscopy, Electron; Microtubules⏎Analysis of serial cross-sections of the Ch
-
0 US$ 24.95 167 pages<|endoftext|>Gate is the national monument of India. Situated in the heart of New Delhi⏎, it was designed by Sir Edwin Lutyens.The monument is inspired by Arc de
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
- L4:3 strongly excites commas after citation fields
- L14:35 strongly suppresses
- L5:16 strongly excites
- L11:154 strongly suppresses commas after citation numbers
- L3:71 strongly suppresses
- L10:335 strongly excites commas in catalog, date, and list metadata
- L9:77 excites commas in citations and name lists
- L0:68 suppresses
- L2:66 excites
- L1:86 excites
- L0:232 suppresses
- L0:2921 suppresses
- L14:51 excites colon that introduces a statement
- L0:146 suppresses
- L0:346 excites comma closing an introductory phrase
- L5:1304 excites
- L0:584 excites
- L15:241 weakly excites
- L6:305 weakly excites
- L0:1596 weakly excites
- L15:82 weakly suppresses
- L10:10 weakly excites
- L1:1464 weakly excites
- L6:12 weakly excites
Feeds later layers
- L17:1405 strongly excites
- L19:726 strongly excites comma inside large numbers
- L17:116 strongly suppresses
- L17:118 strongly excites comma after a clause ending
- L18:1394 strongly excites
- L17:2001 strongly excites
- L18:366 strongly excites comma inside numbers and dates
- L17:585 strongly excites
- L19:2645 strongly excites commas in the middle of a sentence
- L18:2043 strongly excites
- L23:184 strongly excites
- L19:682 strongly suppresses
- L19:1826 strongly excites
- L22:368 suppresses
- L26:236 suppresses
- L17:627 excites
- L27:77 excites
- L19:383 excites commas after a clause or list item
- L27:358 suppresses
- L20:2082 suppresses
- L20:1298 excites
- L21:700 excites comma before a new clause
- L17:2597 excites
- L20:476 excites commas that lead into a continuation
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
- which
- particularly
- especially
- 但是如果
- but
- which
- 特别是在
- however
- SqlDbType
- itol
Suppresses
- ,
- ,在
- -,
- ,⏎⏎⏎⏎
- ,b
- ,
- ,d
- ،
- ,www
- ,则