unclear: pieces of names and acronyms
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. 21 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 fires on a fragment of a longer name, acronym or word, such as the Y in YSN or the CHE in Che Guevara. Its pattern is hard to pin down, and turning it up mostly made the model repeat itself or drift into odd word fragments.
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.
-
-scientific scommunities about your work⏎AIMES, in collaboration with the UK, held five Young Scholar’s Network (YSN) workshops. The YSN goal
-
⏎- Paper: #87743533⏎Excerpt from Essay :⏎Perceptions of Che Guevera⏎PERCEPTIONS OF CHE GUEVARA⏎Che Guevara was born as Ernesto Gue
-
to remove soot from diesel emissions in 1978, just a few years after the famed catalytic converter substrate. But like many environmental technologies, DPFs were initially viewed as an unnecessary luxury since diesel engine manufacturers easily met the
-
Woman's Board of Missions of the Interior, located in Chicago (WBMI), the Woman's Board of Missions for the Pacific, located in San Francisco (WBMP), and the Woman's Board of Missions for the Pacific Islands,
-
⏎Vireos. On occasion, good numbers may be encountered, especially in late⏎August and throughout September. Numbers decline thereafter and by November⏎any Warbling Vireo is rather scarce and late.⏎Warbling Vireo is a
-
offers two different day programs that support the opportunity to gain life-enhancing skills, interact with the community, and gain more independence. Learn more by visiting https://www.lsahomes.org/our-programs/⏎- Help your child incorporate the act
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 just dived in the waterbender benden, and I was in the position of dreebuna (duna) in a benden, and I was in a duna (cage) with a duna b
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
0|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce:v:42:y:2011:i:2:p
-
that cover vast areas of water. Find out what happens when other microbes hitch a ride.⏎Get more reference information on Phormidium pseudpriestleyi.⏎Phormidium pseudpriestleyi isolated from a
-
and S to CA; var. melanocarpa – E to AB and S to NM, AZ, NV and N CA; var. leucocarpa – E to PQ and NB and S<|endoftext|>A Classical African Conception of Womanhood
-
or discuss; as, I reasoned the matter with my friend.⏎9. (v. t.) To support with reasons, as a request.⏎10. (v. t.) To persuade by
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
- L20:2048 excites
- L17:528 excites
- L15:2 excites
- L18:304 excites
- L18:22 excites
- L20:263 suppresses
- L20:912 suppresses
- L14:1949 weakly excites
- L8:1433 weakly suppresses
- L1:1428 weakly excites
- L16:653 weakly suppresses
- L10:116 weakly suppresses
- L15:1707 weakly excites
- L1:2316 weakly excites
- L18:34 weakly excites
- L19:119 weakly suppresses
- L18:846 weakly suppresses
- L15:3035 weakly suppresses
- L14:1452 weakly excites
- L12:226 weakly suppresses
- L14:1810 weakly excites
- L14:229 weakly excites
- L13:2349 weakly excites
- L3:2760 weakly suppresses
Feeds later layers
- L22:224 strongly suppresses
- L23:363 excites
- L22:1393 excites
- L26:539 excites
- L24:166 suppresses
- L25:2507 excites
- L26:386 suppresses
- L27:316 suppresses
- L24:710 excites
- L22:1784 excites
- L24:561 suppresses
- L26:1756 excites
- L25:2443 excites
- L23:404 excites
- L25:246 suppresses
- L24:355 suppresses
- L26:977 suppresses
- L25:593 suppresses
- L23:383 excites
- L27:830 weakly suppresses
- L26:321 weakly suppresses
- L22:1182 weakly suppresses
- L23:133 weakly excites
- L25:1080 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
- 功
- OE
- 鸠
- 酌
- 尸
- ję
- etal
- 号线
- antro
- 分行
Suppresses
- ONGO
- ripp
- xFFF
- Pandora
- ratt
- xFFFF
- :r
- ipple
- ivicrm
- @g