author surnames in citations
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 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 end of a researcher's surname in a citation, such as 'Postman' or 'Pullman', and on 'et' in 'et al.'. It turns strongly negative on the word 'of' right after a researcher's name. Steering gave no consistent topic.
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.
-
strongly reflects the view advocated by Neil Postman (1979) of creating a “thermostatic mechanism” that
-
lpd.⏎How to pronounce daemon His Dark Materials?⏎a demon (/ˈdiːmən/listen)) is a type of fictional being in Philip Pullman's fantasy trilogy His Dark Materials.⏎How do you say mail daemon?⏎So the
-
what happens when Lila meets Razvani and if she survives to tell the tale.⏎I think we need to look out for some more stories by Philip Pullman. Can you recommend any for us?⏎Year 4 Blog⏎We are Miss
-
Quintín Bandera Betancourt (ca. 1834 – 1906), also Quentin Bandera and Quintin Bandera, was a military leader of the Cuban insurrection against the Spanish during the Cuban War of
-
Economics,⏎American Economic Association, vol. 3(2), pages 137-66, April.⏎- Angus Deaton & Olivier Dupriez, 2009. "Purch<|endoftext|>Of Biochemistry and Bel
-
a coalition of health organizations published new guidelines for cervical cancer screening. DCEG research into the benefits of incorporating HPV testing into cervical cancer screening programs (Katki et al., 2011) have informed these revised guidelines. More information
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 lately, and I have been thinking about lately. I have been thinking about lately, and I have been thinking about lately. I have been thinking about lately, and I have been thinking about lately. I have been
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
because we can extract it from bones of woolly mammoths, which date back tens of thousands of years, and make sense of it," explains Nick Goldman of EMBL-EBI. "It's also incredibly small, dense and
-
the way players earn forgiveness are not necessarily congruent with harsh punishments.⏎Research by Michael Wenzel of Flinders University in Australia has illuminated the difference between two conceptions of justice.
-
or pleasure, or even sees that pumpkin bread with chocolate chips bringing someone else happiness, then it can improve a sense of wellbeing,” says researcher Jacqueline Gollan of
-
ia in human culture⏎In the last quarter of the 18th century, Professor Jacob Seydelmann of Dresden developed a process to extract and produce a more concentrated form for use in water
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
- L0:351 suppresses
- L0:1224 suppresses
- L0:178 suppresses
- L0:638 weakly suppresses
- L0:300 weakly excites
- L0:454 weakly suppresses
- L0:1700 weakly excites
- L0:381 weakly excites
- L0:809 weakly suppresses
- L0:3066 weakly suppresses
- L0:2175 weakly excites
- L0:417 weakly suppresses
- L0:780 weakly suppresses
- L0:571 weakly suppresses
- L0:539 weakly suppresses
- L0:960 weakly suppresses
- L0:64 weakly excites
- L0:1534 weakly excites
- L0:747 weakly suppresses
- L0:640 weakly suppresses abbreviation periods after initials
- L0:577 weakly suppresses
- L0:1394 weakly suppresses
- L0:1963 weakly excites
Feeds later layers
- L5:1533 suppresses
- L2:2867 excites
- L7:361 weakly excites
- L22:2578 weakly excites
- L21:1175 weakly excites
- L3:810 weakly excites
- L3:334 weakly suppresses
- L19:1067 weakly excites
- L6:1430 weakly excites
- L15:380 weakly suppresses
- L2:1793 weakly excites
- L18:671 weakly excites
- L4:287 weakly suppresses
- L14:984 weakly suppresses
- L10:2918 weakly excites
- L8:1372 weakly excites
- L15:1169 weakly suppresses
- L9:165 weakly suppresses
- L21:2337 weakly excites
- L12:1325 weakly excites
- L6:1159 weakly excites
- L6:2616 weakly excites
- L6:2348 weakly excites
- L14:2375 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
- evity
- Origin
- ][/
- Favorite
- MethodInfo
- Congress
- ularity
- cerpt
- Covenant
- 无敌
Suppresses
- ands
- 和支持
- ポート
- UBLE
- 实验室
- kan
- andle
- Nd
- nod
- .Support