inside Italian names and places
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. 24 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 middle pieces of Italian names and places, such as 'Pal-azzo Pitti', 'Pier-antonio' and 'Politecnico di Milano'. It goes strongly negative on Italian words in glossary-style lists. Turning it up or down did not make the model's writing noticeably Italian.
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.
-
families. The Medici collection began in the 16th century, first in Palazzo Vecchio, then at the Uffizi and finally in Palazzo Pitti with the foundation of the Accademia del Cimento in 1
-
1416 and initially worked as a Greek scribe. His study under the famous schoolmaster Vittorino da Feltre gave him a mastery of Latin rare among Greek emigres,
-
cent. produced many glorious calligraphic works. Among the best known of these masters were Matteo Contugi, Gianrinaldo Mennio, and Pierantonio S
-
of Kenya (January, 2009), a German canton of Switzerland (March, 2009), and the Piemonte-Toscana region of northern/central Italy (images only, February-March,
-
2⏎Copyright (c) 2009 Silvia TAVERNINI, Daniele NIZZOLI, Giampaolo ROSSETTI, Pierluigi VIAROLI⏎This work is licensed under a Creative Commons Attribution
-
Can self-assembly based technologies offer advantages beyond conventional top-down lithography approaches?⏎A joint effort of the Aalto University of Helsinki, the Politecnico di Milano, and VTT Technical Research Centre of Finland has now demonstrated that it is
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: because of the e-file because of the e- due because of the e- due because of the e-file because because because because because because because because because because because because because because because because because because because because because because because because because because
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the frequency.⏎Deutsch: Fabrik / Español: Fábrica / Português: Fábrica / Français: Usine / Italiano: Fabbrica⏎A factory (previously manufactory) or manufacturing plant
-
by an adjustable valve and a scaven<|endoftext|>Deutsch: Faktor / Español: Factor - Factor / Français: Facteur - Facteur / Italiano: Fattori - Fattori⏎A factor, a Latin word meaning may
-
Is.⏎|Prepared by:||Gordon Rudloff⏎Ohio Department of Agriculture<|endoftext|>Logic & ProofsOpen Learning Initiative⏎Cosa impari in questo corso?⏎Logic & Pro
-
Europe, Snam and TIGF) based in six EU Member States plus two renewable gas producing organisations (European Biogas Association and Consorzio Italiano Biogas). The group shares the vision that renewable and low carbon gas,
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
- L21:711 strongly excites
- L23:352 strongly suppresses
- L23:268 suppresses
- L22:759 excites
- L23:166 suppresses
- L23:322 excites
- L22:461 excites
- L22:700 excites
- L21:266 excites
- L16:1805 excites
- L13:771 weakly suppresses
- L22:915 weakly suppresses digits inside historical years
- L5:110 weakly suppresses
- L6:7 weakly suppresses
- L17:640 weakly excites
- L8:3 weakly excites
- L22:870 weakly suppresses
- L1:1431 weakly suppresses
- L0:2672 weakly excites
- L20:565 weakly excites
- L4:2587 weakly excites
- L21:437 weakly suppresses
- L6:13 weakly excites
- L6:8 weakly suppresses
Feeds later layers
- L25:744 strongly suppresses
- L26:2099 strongly suppresses
- L27:2706 strongly suppresses
- L26:3021 suppresses
- L26:2570 suppresses
- L26:2402 suppresses
- L26:2462 suppresses
- L27:920 suppresses
- L26:2428 suppresses
- L27:2694 suppresses
- L27:2789 suppresses
- L27:1824 suppresses
- L27:2514 suppresses
- L26:1970 suppresses
- L27:2690 suppresses
- L26:1673 suppresses
- L25:2841 suppresses
- L26:637 suppresses
- L27:2018 suppresses
- L27:1699 suppresses
- L26:2479 suppresses
- L26:1495 suppresses
- L26:1126 suppresses the digit 1 in powers of ten
- L25:1849 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
- ữa
- ",__
- such
- icular
- 向
- Acts
- getNode
- ĩnh
- clubs
- 向下
Suppresses
- come
- ris
- proprietà
- seg
- scop
- 对自己的
- rice
- sia
- ecc
- pubb