unclear: mostly weak, Chinese and Greek characters
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. 1 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 barely responds to anything. Its highest hits are on odd spots in Chinese poetry and Greek text, and turning it up or down gave no clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
systems, it also means that pollutants that originate indoors are getting trapped inside.⏎“People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落知多少⏎
-
: spring 眠 : sleep 不 : not 觉 : to feel, wake up 晓 : dawn⏎Meaning for Lien 2 处 : place 闻 : hear 啼 : cry 茑 : bird⏎
-
ْتَّصَوُّف; صُوفِيّ ṣūfiyy / ṣūfī, مُتَصَوِّف mutaṣawwif), various
-
来进行研究,探讨泰国学生在学习汉语助动词中的困难及偏误情况,分析他们产生偏误的原因。我们发现经过基于对比研究设计的汉语助动词课堂教学后,“后测(Post-test)”
-
more than 3,800 among adults 55 to 64 and around 800 among adults older than 65 — they accounted for a smaller proportion of deaths in these populations.⏎Among the eldest two
-
). Complex carbohydrates include all carbohydrates with long chains of monosaccharides (oligosaccharides and polysaccharides).⏎From a chemical point of view, it makes
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 a lot of the time of the first. I need to think about about the second. I need to think about the third. I need to think about the fourth. I need to think about the fifth. I need to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ἔλεγεν αὐτοῖς,⏎τὸ σ
-
’ οὐ διὰ τὸν τόπον τὸ ἔθνος⏎ἀλλὰ διὰ τὸ �
-
(core) city ; Greek ἄστυ asty ) with the associated surrounding area ( χώ
-
�ὔτε γὰπ τὸ χωρίον τοῦτο, ἐν ᾧ τὸ ἰερὸν φυλάττ<|endoftext|>Iron and Stone by
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
- L0:1224 suppresses
- L0:1301 suppresses
- L0:178 suppresses
- L0:1470 weakly suppresses
- L0:2764 weakly suppresses
- L0:97 weakly excites
- L0:1072 weakly suppresses
- L0:1634 weakly suppresses
- L0:2797 weakly suppresses
- L0:1318 weakly suppresses
- L0:30 weakly suppresses
- L0:3003 weakly suppresses
- L0:2460 weakly excites
- L0:489 weakly excites
- L0:894 weakly excites
- L0:220 weakly suppresses
- L0:1387 weakly excites
- L0:1362 weakly suppresses
- L0:2569 weakly excites
- L0:1733 weakly excites the word 'system' or 'systems'
- L0:1168 weakly excites
- L0:1 weakly suppresses
- L0:3005 weakly suppresses
- L0:2164 weakly excites
Feeds later layers
- L7:1693 weakly suppresses
- L7:2369 weakly suppresses comma or 'to' before a quotation
- L19:1660 weakly suppresses
- L15:2788 weakly excites
- L6:500 weakly suppresses
- L13:353 weakly excites
- L4:3018 weakly suppresses
- L5:1245 weakly suppresses
- L11:2725 weakly suppresses
- L12:2973 weakly suppresses
- L25:1472 weakly excites
- L14:1796 weakly suppresses
- L3:2140 weakly suppresses
- L24:1837 weakly suppresses
- L19:2932 weakly excites
- L26:436 weakly suppresses
- L21:1274 weakly suppresses
- L7:1115 weakly suppresses
- L12:2818 weakly excites
- L11:1710 weakly suppresses
- L20:2809 weakly excites
- L25:2540 weakly suppresses
- L5:1293 weakly suppresses 'of' in organisation names like 'Ministry of'
- L6:2455 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
- Sabha
- صب
- thic
- ὴ
- quo
- ThanOrEqualTo
- .Management
- 则是
- 腙
- esis
Suppresses
- antics
- aten
- aversable
- ERAL
- lü
- enberg
- culo
- ழ
- @"⏎
- sson