period after a capital initial
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. 3 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 after a single capital letter, the kind used in abbreviations or initials such as "A. indica", "E. coli" or "M. D. Anderson". Turning it up did not clearly change the written text.
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.
-
for its multitudinous medicinal properties (Tanguturi et al. 2020). Different compounds with biological activity have been reported obtained from A. indica as alkaloids, flavonoids, tr
-
under experimental conditions. We compared weight variation and survival between control groups and mixed groups during three years of monitoring. We found both weight loss and high mortality in E. orbicularis of the mixed groups. This study argues for applying a precaution principle
-
different types of tumors including benign, malignant, primary, metastatic, single and multiple. A study in 2008 by The University of Texas M. D. Anderson Cancer Center revealed that SRS and Whole brain radiation therapy (WB
-
It is also known from Taquina, where specimens were collected in 1914 and the species was recorded commonly in 2010 (C. Olaciregui in litt. 2012), and a population
-
ICU have invoked the re-emergence of colistin, an antibiotic that had rarely been used for decades. However, colistin-resistant strains of K. pneumoniae have been reported in Greek ICUs. In
-
Stinnett was selected because it was out of the rough country along the Canadian River.⏎On July 24th, 1926, A. S. Stinnett and Joe Williams paid W. A. Starnes
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 need to get the answer. I am a student, I need to learn. Okay, this is a problem. So let me try to solve. I am a student. I need to learn. Okay, let's see. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
working to have it patented. Once it becomes available, it could be the future of electronics, from smartphones to wearables, and even electronic vehicles. "[T]his is a proof-of-concept demonstration, and our studies show there are very
-
, Aristotle, Aquinas, Hume, Kant, Mill, Nietzsche, MacIntyre, and Hursthouse, the Bhagavad Gita, Hsun-Tzu, and Confucian works. Literary figures include Sophocles,
-
banks are pre-glacial in origin. The earliest settlement was based on river transport, first by the cottonwood dugout canoes of the Lheidli T'enneh, then by birch bark canoes of fur traders. Around the
-
- 27°C (Ref. 12468); 5°N - 4°N⏎Africa: Lake Barombi Mbo (Ref. 4999, 81260),
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
- L2:263 weakly excites
- L2:1838 weakly excites
- L2:2786 weakly excites
- L2:1165 weakly excites
- L1:1228 weakly excites
- L1:1016 weakly suppresses
- L2:157 weakly suppresses
- L1:2853 weakly excites
- L2:1137 weakly suppresses
- L2:1655 weakly suppresses
- L0:1224 weakly suppresses
- L0:803 weakly suppresses
- L0:1029 weakly excites
- L2:897 weakly excites
- L2:2001 weakly suppresses
- L2:59 weakly excites
- L2:1182 weakly suppresses
- L0:178 weakly suppresses
- L0:2518 weakly suppresses
- L0:1301 weakly suppresses
- L2:1793 weakly suppresses
- L2:803 weakly excites
- L2:375 weakly suppresses
- L2:112 weakly excites
Feeds later layers
- L21:1597 excites
- L4:2604 weakly excites
- L17:573 weakly excites
- L20:1416 weakly suppresses
- L15:2608 weakly excites
- L14:1305 weakly excites
- L18:663 weakly suppresses
- L4:595 weakly suppresses
- L7:2655 weakly suppresses
- L23:2662 weakly suppresses
- L7:765 weakly excites
- L26:2403 weakly suppresses
- L13:1612 weakly excites
- L12:434 weakly suppresses
- L14:1207 weakly suppresses
- L13:128 weakly excites
- L20:404 weakly excites
- L13:2851 weakly suppresses
- L26:745 weakly excites
- L19:2490 weakly suppresses
- L6:2084 weakly excites
- L16:2304 weakly suppresses
- L6:274 weakly excites
- L7:1910 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
- enus
- roy
- realizing
- Figure
- figure
- realizes
- geography
- &:
- stance
- Fig
Suppresses
- illos
- cpt
- 是非
- illo
- inkle
- owners
- icolor
- wipe
- 所以我们
- pard