opening parenthesis 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. 24 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 open bracket right after a name, just before its short form appears, like 'Advanced Plasma Power (APP)'. Turning it up made the model drift toward analysis and algorithm talk.
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.
-
L) — can help predict when athletes should return from sports-related concussions, a multicenter, prospective diagnostic study was launched and led by the Advanced Research Core (ARC) of the Concussion Assessment, Research, and Education (CARE)
-
companies that includes Advanced Plasma Power (APP), is building a full-scale gas plant that will prove this technology is commercially
-
made in laboratories around the world. And they are among the goals of more than 100 horizon-stretching research projects housed in the Advanced Measurement Laboratory (AML) Complex at the National Institute of Standards and Technology (NIST).⏎Completed
-
, it may sound like a pipedream (sorry!) but try telling that to the 20 members of the Young Persons Network who visited Advanced Plasma Power (APP) in Swindon to see for themselves how this dream is becoming a
-
The Australian Government is making it easier for new migrants to learn and improve their English language skills by increasing the number of available hours through the Adult Migrant English Program (AMEP).⏎From today, migrants in Australia will
-
funding,” Turner said.<|endoftext|>This article touches on cultural, linguistic, and technical issues that arise in video recording stories of Deaf individuals who use American Sign Language (ASL). Whether you are Deaf or hearing, documenting ASL interviews presents
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: if a business is supposed to be a "center of a city," then the center of a city is a center of a center, and a center of a center of a center, and so on, with a cycle of three or more
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
will make your list alphabetical A to Z. (Descending will reverse the list from Z to A).⏎There is an option of telling the program that your list has a header
-
affectionately known as the people’s memorial, until our memorial there had been no truth and reconciliation process for this heinous piece of our shared history. Sankofa Day is the first steps in the truth, healing and reconciliation process done for
-
ues to Contaminate Surface Water and Drinking Water in the United States⏎Banned in the European Union and clearly linked to harm to wildlife and potentially to humans, the pesticide atrazine provides little benefit to offset its risks. In 20
-
in a variety of different conditions. One notable case of such similarity, adenosine (ADO), shares the purine skeleton and oxygenated pendant groups. Together, the two molecules look like this:⏎Adenosine
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
- L21:3022 strongly suppresses
- L22:178 strongly suppresses
- L22:441 strongly suppresses
- L23:2946 excites
- L21:915 suppresses
- L23:2635 excites
- L22:672 suppresses
- L22:1545 suppresses
- L20:2900 excites
- L21:337 excites
- L20:2780 suppresses
- L21:1046 excites
- L22:985 excites
- L17:2712 excites
- L23:1358 suppresses
- L21:1602 excites
- L22:2394 suppresses
- L21:2651 weakly suppresses
- L22:411 weakly suppresses
- L20:2808 weakly excites
- L21:90 weakly excites
- L18:51 weakly excites
- L20:1200 weakly suppresses 'and' in 'more and more' doubling
- L18:2455 weakly excites
Feeds later layers
- L25:568 strongly suppresses
- L27:652 strongly suppresses
- L27:709 suppresses
- L26:411 suppresses
- L25:315 excites
- L25:1361 suppresses
- L26:517 excites
- L27:231 excites
- L27:173 suppresses
- L27:203 weakly suppresses
- L26:1164 weakly excites
- L26:484 weakly suppresses
- L25:1014 weakly suppresses
- L27:767 weakly excites
- L27:1025 weakly excites
- L27:919 weakly suppresses
- L27:1042 weakly suppresses
- L25:2761 weakly suppresses
- L27:1292 weakly excites
- L27:548 weakly suppresses
- L25:754 weakly excites
- L26:2487 weakly excites
- L27:618 weakly excites
- L26:220 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
- アル
- analyzer
- (api
- articulate
- ascii
- artic
- 닝
- alg
- 算法
- agan
Suppresses
- A
- A
- a
- →A
- *A
- :A
- _a
- .a
- <A
- _A