period inside an abbreviation
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. 15 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 full stop in short forms like E. coli, Univ. Arizona, B.C. or U.S., where the dot does not end a sentence. It is quiet or negative on question marks and full stops that really do end a sentence. Turning it up gave rambling text that did not clearly reflect 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.
-
the dilemma than Boone County, where major urban streams such as Flat Branch, which recently tested for seven times the safe limit for E. coli, remain unprotected.⏎The county has 94 treatment systems that would have
-
er.2008 May 27⏎Credit: Phoenix, HiRISE, NASA, JPL-Caltech, Univ. Arizona⏎Explanation: The Phoenix lander's footpads are about the size
-
attempting to counter the claim that articulation has a one-to-one correspondence with burial. I was responding to Hovers, E. and Belfer-Cohen, A. “Insights into early mortuary
-
, observance of a pause in between was necessary. This we have already explained in the introduction to Surah Al Muddaththir; and in E. N. 5 of Surah Al-Muzzammil also we have explained
-
. 3-4.5m high. These are covered in deciduous woodland. On the east side the camp defences consist of an inner bank C. 0.5-1m high on inside fence, and well-preserved
-
This dogma has been described in the Vyāsabhāsya on the Yogasutras of Patañjali (200 B. C.). The word ‘ekabhavika’ has been introduced and discussed here
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 I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
bullets, but few⏎of the horses escaped. Tender-hearted gunners took time out to<|endoftext|>Is PET Plastic Safe for Food and Beverage?⏎Is PET plastic safe?
-
eventual climate impacts<|endoftext|>Transforming Canada’s Food Guide: What’s missing?⏎In the aim to provide a full diet, what does Canada’s Food Guide leave out? Take
-
essential. This comprehensive guide aims to provide a clear understanding of what a bilan comptable entails, including its key components such as assets, liabilities, and equity. By delving into the importance of the bilan comptable for financial analysis and
-
minster Township, and it consisted of 105 acres of land, part of<|endoftext|>How do you bring an endangered plant species back from the brink of extinction? The answer might be found in zoo animals.⏎That’s the inspiration for Chicago Bot
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
- L14:422 excites periods after initials and abbreviations
- L14:155 excites sentence-ending periods in prose
- L14:51 excites colon that introduces a statement
- L4:7 excites
- L5:155 suppresses
- L5:168 suppresses
- L11:132 excites period ending an opening topic sentence
- L9:2447 excites
- L6:305 weakly suppresses
- L8:541 weakly excites
- L9:7 weakly excites
- L6:250 weakly excites
- L7:64 weakly suppresses
- L14:634 weakly excites
- L2:9 weakly suppresses sentence-final periods vs dots in code
- L4:1546 weakly suppresses
- L10:612 weakly excites full stop ending boilerplate or a title
- L8:143 weakly suppresses
- L8:740 weakly excites
- L13:53 weakly excites period ending a claim before elaboration
- L13:95 weakly excites
- L0:2602 weakly excites
- L12:5 weakly excites
- L10:921 weakly excites
Feeds later layers
- L17:1772 suppresses
- L16:1545 suppresses
- L21:1086 suppresses end of sentence before a claim or statistic
- L20:473 suppresses
- L17:1105 suppresses
- L17:347 suppresses full stops ending informative sentences
- L19:384 suppresses
- L19:484 suppresses
- L25:198 suppresses
- L24:353 suppresses decimal point inside a number
- L17:231 excites
- L16:489 excites
- L17:990 suppresses
- L23:179 suppresses
- L18:227 excites
- L20:822 suppresses
- L27:133 suppresses
- L16:339 suppresses
- L27:1637 suppresses
- L20:296 suppresses
- L22:282 suppresses period then line break in lists
- L20:343 suppresses
- L20:940 suppresses
- L22:67 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
- 缮
- .&
- МА
- wind
- об
- vron
- ma
- 肃
- RequestMethod
- .`
Suppresses
- ernet
- 不过
- -*-
- rän
- Each
- Also
- ään
- moreover
- ⏎→→⏎
- ⏎⏎