periods after initials and abbreviations
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 full stop that follows a single capital letter or a short abbreviation, like the dot in 'D. C.', 'E. coli', or 'J. Murray'. It doesn't clearly change the model's writing when turned up, so the steering results don't 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
-
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
-
in Washington, D. C. (Image via Zach Montellaro on Flickr, CC BY-NC
-
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
-
from the arthropod realm are summarized and their potential use in allergy diagnostics discussed.<|endoftext|>GRĪW, a Middle Iranian word (in Manichean Mid. Pers. script gryw, rarely gryyw; in Sogd
-
LANGUAGE TEACHING AND RESEARCH|⏎|DOI - pysyväislinkit|⏎|Tila||Julkaistu - 1 syysk. 2021|⏎|OKM-julkaisutyyppi
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 we have, we have, we we, we, we we, we we we, we we we, we we we, we we we, we we we, we we we, we we we, we we we, we
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
-12, college, and post-college.⏎To view this DRM protected ebook on your desktop or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an
-
solving of nonmathematical problems as mathematical problems.⏎To view this DRM protected ebook on your desktop or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an
-
be God, the all-loving creator who meets every human need. In Psalms we read, “Give thanks to the Lord, for he is good; his love endures forever. Let the redeemed of the Lord tell their story. .
-
Animal Adaptations". Please register for the second session through Eventbrite as well, if you are interested in attending both: https://scisatadapt2.eventbrite.ca⏎Please contact us if you have any questions & we hope to
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
- L4:7 excites
- L2:9 suppresses sentence-final periods vs dots in code
- L8:143 suppresses
- L5:155 suppresses
- L5:168 suppresses
- L1:130 suppresses
- L6:305 suppresses
- L6:972 excites "well" after "how"
- L5:1304 excites
- L4:1546 suppresses
- L11:132 excites period ending an opening topic sentence
- L6:250 weakly excites
- L2:1473 weakly excites
- L4:300 weakly excites
- L5:454 weakly suppresses
- L5:125 weakly excites "of" after all three, both, ten
- L0:36 weakly excites
- L9:7 weakly excites
- L1:1 weakly excites
- L11:1981 weakly suppresses
- L10:33 weakly suppresses
- L9:240 weakly excites
- L3:272 weakly excites
- L0:1042 weakly suppresses "This" starting a labelled sentence
Feeds later layers
- L15:344 excites
- L15:403 excites period inside an abbreviation
- L17:642 excites
- L16:432 excites period after month or initial
- L17:347 excites full stops ending informative sentences
- L17:1770 excites
- L18:791 excites
- L17:1772 excites
- L15:241 excites
- L15:117 suppresses
- L15:313 excites decimal point inside numbers
- L19:831 excites
- L16:1545 excites
- L18:2497 excites hyphen inside compound words
- L17:1622 suppresses
- L27:133 excites
- L27:12 excites
- L18:713 excites
- L17:1155 excites opening parenthesis after a sentence
- L20:962 excites decimal points and dots in numbers
- L20:505 weakly excites
- L21:1237 weakly excites
- L22:67 weakly excites
- L16:797 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
- ugal
- мот
- 間
- рай
- alternative
- let
- 朋友们
- 례
- ев
- Nature
Suppresses
- MainThread
- ��
- lambda
- abilia
- egin
- backs
- nable
- ifies
- �
- yor