opening bracket before an acronym
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. 19 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 an opening bracket that comes right after a full organisation or term name, just before its short form, such as 'Centers for Disease Control and Prevention (CDC)'. It goes strongly negative on letters inside an acronym, and steering did not show a clear link to this.
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.
-
:21 "I spoke to you in your<|endoftext|>How’s your facility looking this flu season? According to the U.S. Centers for Disease Control and Prevention (CDC), this year’s flu season is shaping up to become the worst the United
-
sensitivity, says Cowley: "This work really does point to pursuing that as a<|endoftext|>(NaturalNews) Ohio farmers are more reluctant to plant genetically modified organisms (GMOs) in their fields than their counterparts in other Midwestern states,
-
death for American teens. There were more than 3,000 teen fatalities in 2010, according to the National Highway Traffic Safety Administration (NHTSA).⏎- In 2013, Ford Driving Skills for
-
1.7 million in 2007, or nearly 4% of the population, according to new figures from the Centers for Disease Control and Prevention (CDC).⏎That's up from about three million in 1971 and
-
o’s Dissertation as a PDF […]⏎The World As It Could Be Curriculum and Resource Guide for Teachers & Organization Leaders is now available through the Education Resources Information Center (ERIC) at http://www.eric.ed.gov/ERICWebPortal/detail?
-
side bar!⏎How is Computer Science different from Computer Information Systems and other IT disciplines?⏎A frequently-asked question concerns the distinction between CS and Computer Information Systems (CIS). While the specific details of<|endoftext|>In 1919,
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: $22$ $22$ $22$ $22$ $22$ $22$ $22$ $22$ $22 $22$22$22$22$
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ography. What’s changing for TSMC at 20nm is that the company is adopting what’s called double pattern/double etch (2P2E).⏎(The below image shows the same cell area built with single exposure, double
-
comes with some significant costs.⏎For nearly a decade, TSMC, GlobalFoundries, and Intel have collectively relied on argon-fluoride (ArF) lasers to etch microprocessor wafers. These lasers generate light
-
. In this disease, there appears to be a specially associated member of the family of human endogenous retrovirus-W known as MS-associated retrovirus (MSRV).⏎During pregnancy in
-
Missions for the Pacific, located in San Francisco (WBMP), and the Woman's Board of Missions for the Pacific Islands, located in Hawaii (WBMPI). The three mainland boards cooperated with the maleled American Board of Commissioners
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
- L17:2755 suppresses
- L17:2502 suppresses opening parenthesis before an abbreviation
- L17:2032 suppresses
- L14:2419 excites
- L17:1307 excites parenthesis introducing an acronym
- L18:2326 excites
- L18:1031 excites
- L1:2637 excites
- L16:1605 excites
- L18:2674 weakly suppresses
- L17:1404 weakly suppresses
- L13:370 weakly suppresses
- L16:1849 weakly suppresses
- L16:225 weakly excites
- L17:621 weakly excites
- L17:622 weakly excites
- L3:2032 weakly suppresses
- L13:2248 weakly excites
- L12:2252 weakly suppresses
- L13:532 weakly suppresses
- L13:821 weakly suppresses
- L12:227 weakly suppresses
- L16:1048 weakly suppresses
- L14:2617 weakly suppresses
Feeds later layers
- L21:857 strongly suppresses
- L20:1080 strongly suppresses opening paren before an acronym
- L20:2479 strongly suppresses
- L21:2345 strongly suppresses
- L23:404 suppresses
- L21:144 excites
- L25:446 suppresses
- L21:1443 suppresses abbreviations in parentheses
- L27:266 suppresses
- L21:1291 suppresses
- L25:652 suppresses
- L20:824 suppresses unclear: opposite pairs and citation digits
- L25:414 excites
- L21:362 suppresses
- L25:735 suppresses
- L20:1675 suppresses
- L21:3012 suppresses
- L23:349 suppresses comma after author surname in citations
- L20:1306 suppresses
- L20:912 suppresses
- L20:1049 suppresses
- L21:1014 excites
- L26:394 suppresses
- L20:389 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
- !⏎⏎⏎⏎⏎⏎
- .\"
- 詹姆
- twe
- ihad
- ) ⏎
- 就应该
- \",⏎
- HorizontalAlignment
- legality
Suppresses
- CC
- cc
- EE
- EA
- FS
- req
- Req
- eq
- FB
- cb