blank spot before a run-on number
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. 14 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 empty gap just before a number that has been attached to the end of a sentence or header, such as a footnote number, a comment count, or a time after a date. Turning it up made the model produce repetitive lists and labels, which doesn't clearly match this pattern.
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.
-
a good footnote and give you a place to start if you develop constant and persistent joint pain, but they are not predictive and certainly not diagnostic.⏎| Related: 9 Everyday Surprises When RA Enters Your Life |⏎Many of these risk factors
-
games [The Daily Journal]<|endoftext|>Should You Ask Your Children to Apologize?By Craig Smith | September 23, 2016 | 0 comments⏎When kids say sorry, are they learning a lesson or just par
-
evidence indicating that living systems may be designed.⏎The Kansas Citizens for Science has recently urged 1 you to take the former path, to teach only naturalism. 2 They have asked you to reject the Science Education Standards adopted by the Kansas State
-
, turn by turn directions, UI, User interface, video chat, Virtual reality, Virtual reality glasses⏎January 18, 2012 • 11:07 am
-
5, for example, a separate team developed a tiny, $0.20 temperature sensor that didn’t require a battery or constant source of wired electricity. 3D printing isn't new either, and already, researchers can print out human
-
Struggle-For-Power-In-The-994323.html⏎"The Struggle for Power in the Ussr" StudyMode.com. 05 2012. 05 2012
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:,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Arabian lava field from the International Space Station:⏎Harrat Khaybar, Saudi Arabia is featured in this image photographed by an Expedition 16 crewmember on the International Space Station. The western half of the Arabian
-
Exercise 6.5⏎In NCERT Solutions for Class 10 Maths Chapter 6 Exercise 6.5, we have discussed the
-
body feels no pain.⏎Let your child know that these cremated remains are placed in a⏎container called an urn and returned to the family. If cremation⏎has already taken place and the container picked up, you may want⏎to show
-
rules, where⏎are they all sitting? Can you find all the possible ways?⏎What is the best way to shunt these carriages so that each train⏎can continue its journey?⏎If you split the square into these
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
- L9:7 excites
- L8:30 excites blank slot right before a number
- L6:0 excites
- L1:1 excites
- L10:33 suppresses
- L12:5 excites
- L3:3 suppresses
- L7:961 excites
- L3:691 weakly excites
- L1:108 weakly suppresses
- L8:4 weakly excites
- L0:2098 weakly excites
- L8:16 weakly suppresses
- L8:10 weakly excites
- L0:1247 weakly suppresses
- L0:324 weakly suppresses
- L8:14 weakly excites
- L6:1263 weakly suppresses
- L8:2 weakly excites
- L3:264 weakly suppresses
- L6:4 weakly suppresses
- L7:63 weakly excites
- L6:16 weakly excites
- L7:122 weakly excites
Feeds later layers
- L22:372 excites empty spot before a number or "3D"
- L18:106 excites
- L16:1939 excites line break mid-sentence in wrapped text
- L19:781 excites
- L21:173 excites
- L17:359 excites
- L25:118 excites
- L16:822 excites
- L15:123 excites
- L23:77 excites the space right before a number
- L17:23 excites
- L17:1787 excites
- L18:429 excites
- L20:1067 excites
- L27:598 excites
- L24:130 excites line break at the end of a line
- L22:1744 excites
- L21:189 excites document break token mid-sentence
- L19:581 excites blank slot before a number
- L23:35 excites
- L15:122 excites
- L15:1870 excites
- L19:182 suppresses
- L18:2153 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
- undry
- 本报讯
- issued
- occan
- .addComponent
- gay
- 让用户
- 霸王
- ITHER
- 곱
Suppresses
- leshoot
- Wer
- entimes
- ification
- нтер
- atoon
- rze
- irling
- 为主
- areth