inside multi-part proper names
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 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 responds to the middle or later part of a longer name, most often 'Rudolf' in names like Rudolf Virchow or Rudolf Abel. Turning it up did not visibly change the model's output in a clear way.
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.
-
a single almond inside. Whichever guest gets the almond in his pudding is guaranteed good luck for the coming year. Children leave out glasses of milk for the Juulnisse, which are elves that live in peoples' homes, and for J
-
that caused tuberculosis prompted von Ruck to return to Germany to work under Koch. He spent about eight months in Koch's Hygienic Laboratory and at Professor Rudolf Virchow's Pathological Institute. Von Ruck was present on 2
-
Benwell, Tyne & Wear)⏎Wall: W (7) to Onnvm (Halton Chesters, Northumberland)⏎The Romano-British name
-
255 files?⏎File type specification:⏎The 255 file extension is associated with the Gimp (also known as GNU Image Manipulation Program), a picture image editing tool for
-
isteddfod is a common way to celebrate St David's Day, particularly for schoolchildren.⏎Recent Google doodles have also included tributes to Heinrich Rudolf Hertz, film critic Francois Truffaut and Charles Dickens, who was born
-
by conventional treatments.⏎The concept of the minimum dose is similar to Arndt-Schultz law (named after the pharmacologist Hugo Schultz and the psychiatrist Rudolf Arndt, who formulated the law).⏎“Weak stimuli accelerate vital activity,
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 have been thinking about the relationship between the two things, and I have been thinking about the relationship between the two things. Maybe the two things are related, and I have been thinking about the relationship between the two things. Maybe the two
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
decades to come,” said Mr McBride.⏎HBRC is working with Maungaharuru-Tangitū Trust about the possibility of the reconnection of
-
Kingfisher (Alcedo atthis) was taken by KM Sim on 29th December 2012 at the Japanese Garden in Jurong.⏎Note that the kingfisher is in its comfort mode, first scratching its
-
in late May of that year, Major General George B<|endoftext|>Jakarta, Indonesia (CNN) -- The death toll from recent eruptions of Indonesia's Mount Merapi volcano has climbed to 273, government officials said Wednesday.⏎The toll
-
.⏎* Malaysian Palm Oil Board (MPOB),⏎6 Persiaran Institusi, Bandar Baru Bangi,⏎43000 Kajang, Selangor, Malaysia.<|endoftext|>Available for download after purchase.⏎This book
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
- L0:1301 suppresses
- L0:178 suppresses
- L0:1224 suppresses
- L0:571 weakly excites
- L0:2159 weakly suppresses
- L0:2173 weakly excites
- L0:310 weakly suppresses
- L0:1127 weakly excites
- L0:434 weakly excites
- L0:1994 weakly suppresses
- L0:2627 weakly suppresses
- L0:1168 weakly excites
- L0:753 weakly excites
- L0:2696 weakly suppresses
- L0:1001 weakly suppresses
- L0:2114 weakly suppresses
- L0:1304 weakly suppresses
- L0:644 weakly suppresses
- L0:1502 weakly suppresses
- L0:802 weakly excites
- L0:63 weakly suppresses
- L0:1116 weakly suppresses
- L0:1 weakly suppresses
- L0:386 weakly suppresses
Feeds later layers
- L9:915 weakly suppresses
- L8:2308 weakly suppresses
- L8:819 weakly excites
- L21:1219 weakly excites
- L22:2645 weakly excites
- L24:666 weakly excites
- L7:1456 weakly excites
- L21:2985 weakly excites
- L26:2834 weakly excites
- L3:2517 weakly excites
- L5:168 weakly excites
- L7:910 weakly excites
- L5:202 weakly excites
- L15:2261 weakly suppresses
- L10:2328 weakly suppresses
- L13:400 weakly suppresses
- L7:2915 weakly excites
- L8:1043 weakly suppresses
- L17:2194 weakly suppresses
- L21:1677 weakly suppresses
- L17:71 weakly excites
- L3:3051 weakly excites
- L24:2709 weakly excites
- L26:2820 weakly excites
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
- blers
- OOSE
- Cyrus
- môi
- bolt
- iconductor
- edly
- ores
- entifier
- anders
Suppresses
- 间
- erto
- agi
- izu
- uten
- uch
- adv
- ute
- ab
- ugl