comma after author surname 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. 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 comma or initial right after a person's surname in reference lists, such as 'Rudman, W.B.' or 'Bullock, B.A.'. Turning it up made the model write garbled text full of single letters and unusual words, so the steering output tells us little about this neuron.
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.
-
a description of a new Elysia species. Bull. Mar. Sci. 30(1): 54-79.⏎Rudman, W.B., 2002 (October 7) Elysia
-
wise, Incorporated.<|endoftext|>Identifying time lags in the restoration of grassland butterfly communities: a multi-site assessment⏎Woodcock, B.A.; Bullock,
-
apua #PapuaNewGuinea. Save them by changing your weekly shop #Boycott4WildlifeTweet⏎Leary, T., Wright, D., Hamilton, S., Singadan, R., Menzies, J
-
LANGLEY, M. A. Crown 8vo. 65. ANATOMY⏎AND PHYSIOLOGY FOR NURSES. By D. C. KIMBER. 8vo. 10s. net
-
, 2008⏎Publication Date: May 16, 2008⏎Citation: Davis, L.M., Spackman, E. 2008. Do crocodilians get the flu?
-
3, “Pharmacotherapy of Bipolar Disorder” (pp. 37–47)⏎Preston, J. D., O’Neal, J. H., & Talaga, M. C. (201
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: the possible connection between the internal and the external, and the internal and the central. As I am 5-6-7-8-9-10-11-12-13-14-15
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Additionally, going all day without eating and having dinner at night can affect the relationship between insulin and blood sugar. This can put you more at risk for Type 2 diabetes. This is one more reason that avoiding food all day and eating all
-
iales, family Anaplasmataceae; Dumler et al.,2001), systematically split into 6 supergroups, lettered from A to F (Lo et al., 2002). Those in supergroups
-
The so-called diabetes epidemic has made headlines, and for good reason. The number of cases of type 2 diabetes (the most common type) is exploding, and many more people are
-
A buildup of ketones can cause a dangerous condition called ketoacidosis. Your healthcare team may ask you to check for ketones, especially if you have type 1 diabetes.⏎When to check for ketones⏎Ketones in your blood
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
- L20:912 strongly suppresses
- L21:306 strongly suppresses
- L22:1519 strongly excites
- L21:362 strongly excites
- L18:1962 strongly excites
- L14:1810 excites
- L20:2317 suppresses
- L20:3046 excites
- L19:1011 excites
- L19:2801 suppresses
- L19:2585 excites
- L22:2855 excites digits inside citation and DOI strings
- L19:2805 excites
- L22:1223 excites
- L18:1217 excites
- L21:1443 excites abbreviations in parentheses
- L19:2663 suppresses
- L18:2983 suppresses
- L20:1675 excites
- L22:930 excites
- L17:2262 suppresses
- L19:1829 suppresses opening bracket before an acronym
- L22:994 suppresses
- L16:2181 suppresses
Feeds later layers
- L26:539 strongly suppresses
- L26:386 strongly excites
- L25:2507 suppresses
- L27:316 excites
- L24:166 excites
- L24:355 excites
- L25:264 excites
- L27:468 excites
- L24:1731 excites
- L26:278 excites
- L26:411 excites
- L26:301 suppresses
- L27:817 suppresses
- L27:295 excites
- L25:1080 excites
- L24:213 suppresses
- L27:470 suppresses
- L26:977 excites
- L27:768 excites
- L25:1851 weakly excites
- L25:735 weakly suppresses
- L27:945 weakly suppresses
- L24:710 weakly suppresses
- L25:246 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
- c
- S
- s
- d
- C
- m
- M
- e
- r
- t
Suppresses
- sms
- erah
- mah
- ocz
- anter
- pbs
- /hr
- pz
- cms
- edException