closing parenthesis after a place or name
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. 5 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
The neuron fires on a closing bracket that follows a short name or place, such as '(Minnesota)' or '(French)'. It is mostly quiet elsewhere and turning it up did not produce a clear theme.
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.
-
holds a flag in both hands as he glances forward. The sculpture is signed "Gutzon Borglum 1929 (illegible) AKUNST FDY NYC". The back of the base is inscribed:
-
will be characterized with quantum optical metrology techniques. The second objective is to enhance the emission brightness and entanglement quality through integration<|endoftext|>UC DAVIS (US) — After a live videoconference with a specialist, rural emergency room physicians are more
-
(Minnesota) for three years before her marriage to Dr. Will Kennicott, of G
-
7 (2013) [Refereed Article]⏎Infectious diseases of wildlife reduce population size and
-
7 x 66 mm|⏎|Printer||American Bank Note Company (U.S.A.) – ABNC|⏎|Signatures||Ernesto (Che) Guevara Serna & Rolando Díaz Astaraín|⏎|
-
ankofa Day™, an annual national memorial in London’s Trafalgar Square to mark this day, celebrate the great Ayitian (Haitian) Revolution and remember and honour the victims of the Transatlantic Human Trafficking and
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 am a 3D model of a 3D in the shape of a cube in the 3D space. My name is in the shape of a cube. Let me know where I am. I think I can take a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Glass” and their use of geometry as speculative and imaginative.⏎In trying to find more about this term, it appears that sadly the intelligent design crowd has laid claim to it. The most I could find (in an admittedly short search) was this
-
the French parliament and supreme court declared that claims to the French throne could not be passed through women. The French Estates chose Philip of
-
analysis is trustworthy. If the analysis is bulletproof, the next question is whether the evidence is substantial enough to claim a discovery. And if a discovery can be claimed, the final question is what, exactly, has been discovered? Scientists can have
-
the northerners, but two of the invading peoples settled on the coast of Palestine, between Gaza and Mount Carmel. The attempted invasion ended Egyptian pretensions to a Syro-Palestinian hegemony.⏎Two more years of peace ensued,
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
- L4:1981 weakly excites
- L4:516 weakly suppresses
- L2:911 weakly suppresses
- L4:955 weakly excites
- L3:2787 weakly excites
- L3:57 weakly excites
- L1:118 weakly excites
- L4:2429 weakly excites
- L0:2822 weakly suppresses
- L2:2319 weakly excites
- L2:1937 weakly excites
- L4:2463 weakly excites
- L1:2117 weakly suppresses
- L4:2632 weakly excites
- L3:2989 weakly excites
- L4:775 weakly excites
- L0:2554 weakly suppresses
- L3:1949 weakly excites
- L2:2904 weakly suppresses
- L3:2409 weakly excites
- L3:2128 weakly excites
- L4:2143 weakly suppresses "in some" opening a qualifier
- L3:503 weakly suppresses
- L2:2581 weakly excites
Feeds later layers
- L9:828 suppresses
- L13:3016 excites
- L6:536 suppresses
- L8:2060 weakly suppresses
- L18:2842 weakly excites
- L9:2384 weakly excites
- L16:1207 weakly excites
- L6:1713 weakly suppresses
- L23:1713 weakly suppresses
- L11:2581 weakly suppresses
- L7:2906 weakly excites
- L19:441 weakly excites
- L22:578 weakly suppresses
- L13:1991 weakly suppresses
- L7:2211 weakly excites 'their' after work and manage verbs
- L18:974 weakly excites
- L6:1698 weakly suppresses
- L14:2167 weakly excites
- L12:821 weakly suppresses
- L24:1221 weakly excites
- L25:1786 weakly excites
- L12:1710 weakly excites
- L18:638 weakly excites
- L8:115 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
- {"
- {'
- ges
- ixer
- ESC
- drop
- {{
- 坏
- nomin
- 大宗
Suppresses
- ford
- ข
- ью
- unge
- .idea
- -chart
- TimeInterval
- =post
- 也不要
- Manning