the name fragment 'Bon' (Bonn, Bonaparte)
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. 23 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 letters 'Bon' at the start of place and person names, most often 'Bonn' in Germany and 'Bonaparte'. It is not a general concept neuron, and turning it up or down did not give clear, related 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.
-
said during the United Nations Climate Change Conference in Bonn, Germany.⏎Speaking at the conference, China’s top climate negotiator X
-
s that came from Saint-Chamond (Loire). He left school of fine arts and architecture of Lyon with honours gained thanks to a silver medal in Bonnefond painting course. After a some year stay in Paris where he became
-
the University of Bonn in Germany and colleagues. The team created a three-dimensional interference pattern by shining
-
en, as well as an inset at bottom left showing two islands discovered by Cook and Marion du Fresne. One of the few maps in the volume drawn by Bonne's son. Engraved by Andre. Condition: A nice impression with
-
traditionally had close economic and cultural relations with both these neighboring regions. Agreeing with the Bon account, they explain that this master, once in Zhang-zhung, combined many
-
Haeger's novel Courtesan about Diane de Poitiers and Henri II. From 22 June to 7 July 1815, Bonapartists considered Napoleon
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: if we look at the total cost of the product in the market, we could potentially instead of having a total cost being an indicator, instead using the total cost that is going into the product's production for the product’s market being produced,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Antalya-Türkiye’de doğdu. Lisans öğrenimini MSÜ Deniz Harp Okulunda tamamlamıştır
-
⏎$3 million grant from the Bill & Melinda Gates Foundation will fund development⏎by Eric Klopfer⏎MIT Education
-
all remained in place to commemorate the British leaders who championed Jewish rights to a homeland.⏎According to Robert Fulford, a journalist on the Toronto newspaper the Globe & Mail, within a short walk in Jerusalem there are streets called Benjamin Disraeli
-
7 to 1947. It is connected to the museum at its top and bottom. © John Grade⏎The new sculpture at the Museum of History and Industry in Seattle
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
- L21:1798 suppresses
- L20:2799 excites
- L20:1750 weakly excites
- L0:1139 weakly suppresses
- L15:1709 weakly excites
- L10:426 weakly excites
- L17:1686 weakly excites
- L11:2349 weakly excites
- L16:2035 weakly excites 'of' after a noun phrase
- L0:1767 weakly excites
- L20:669 weakly suppresses
- L8:14 weakly excites
- L12:2602 weakly suppresses
- L5:2820 weakly excites
- L22:151 weakly suppresses
- L21:977 weakly excites
- L19:1944 weakly excites
- L14:2463 weakly excites
- L10:2479 weakly excites
- L5:75 weakly excites
- L18:1475 weakly suppresses
- L2:874 weakly excites
- L17:3038 weakly excites
- L22:1221 weakly excites
Feeds later layers
- L26:2748 weakly excites
- L25:1505 weakly excites
- L24:478 weakly excites word fragment inside a longer word
- L24:967 weakly excites
- L24:1660 weakly excites
- L27:1119 weakly excites
- L25:674 weakly excites
- L27:219 weakly excites
- L27:1437 weakly excites
- L24:298 weakly excites 'would' in hypothetical or past statements
- L27:401 weakly excites
- L24:2504 weakly excites
- L27:693 weakly excites
- L26:1516 weakly suppresses
- L24:697 weakly suppresses
- L25:2359 weakly excites
- L26:3004 weakly excites
- L24:2802 weakly suppresses
- L26:952 weakly suppresses
- L24:1678 weakly suppresses
- L25:117 weakly excites
- L25:1172 weakly suppresses
- L24:2974 weakly suppresses
- L24:1084 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
- return
- ized
- (return
- >Error
- nde
- attach
- close
- return
- ated
- handleError
Suppresses
- lector
- ajaran
- стве
- aidu
- razier
- ERENCE
- ollo
- цию
- acock
- cabo