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. 25 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 lights up on the opening bracket that comes right after a full name, just before its short form, like the "(" in "Multi Stage Flash (MSF)". Turning it up made the text wander into unrelated topics, so the steering results don't clearly show this pattern.
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.
-
have their influence on one another visually mapped. This methodology was named Interpretive Structural Modelling (ISM). Over the next 20 years, the use of ISM to
-
The unused funds — called Universal Service and Access Funds (USAFs) — are public funds that aim to overcome this inherent commercial bias
-
and teaching your peers and their non-scientific scommunities about your work⏎AIMES, in collaboration with the UK, held five Young Scholar’s Network (YSN) workshops. The YSN goal
-
ise, so fresh water evaporates and the brine is left behind. Cooling and condensation are then used to obtain fresh water for consumption.⏎Multi Stage Flash (MSF), the most common thermal technique accounting for around 27% of
-
as the National Qualification for Headship (NPQH) in the United Kingdom and the (Interstate School Leaders Licensure Consortium (ISLLC) standards in the United States. Distributed patterns of leadership have developed
-
been analysed over three years (2004-2006). The studied lakes, Ca' Morta (CM) and Ca' Stanga (CS), have a common origin and are contiguous, but are of different ages and
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 can only be the person who is the only one who can do the action. I am the only one who can do the action, and I am the only one who can do the action. I am the only one who can do
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to make the Internet safer. That’s why early in that year, they deployed Netscape Navigator 1.1 giving birth to the Secure Sockets Layer Protocol (SSL in short). As we now know:⏎Netscape’s goal was to
-
have now made history together on multiple occasions. Most recently, the duo completed the first crewed mission conducted by a private spaceflight firm to the International Space Station (ISS).⏎Now, NASA and SpaceX have their sights set on loftier goals.
-
that can be performed in space. The winner/winning team’s experiment will be conducted aboard the International Space Station (ISS), making it the largest science lesson streamed live for the world to watch on
-
risk factors. Mark Goldstein, 79,<|endoftext|>Sea surface temperature⏎Sea surface temperature (SST) is the water temperature at the surface. In practical terms, the
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
- L24:438 strongly suppresses
- L24:1401 strongly suppresses
- L22:2583 excites
- L23:2424 excites
- L22:72 suppresses
- L24:2510 suppresses
- L21:415 excites
- L22:110 excites
- L22:705 weakly excites
- L21:671 weakly excites
- L21:883 weakly suppresses
- L22:1337 weakly excites
- L20:357 weakly suppresses
- L24:2780 weakly excites
- L4:1389 weakly excites
- L23:2111 weakly excites
- L23:895 weakly suppresses
- L17:1070 weakly excites
- L22:533 weakly suppresses
- L0:1635 weakly excites
- L0:1097 weakly suppresses
- L3:2171 weakly suppresses
- L20:2381 weakly suppresses
- L2:1008 weakly excites
Feeds later layers
- L26:1645 strongly excites
- L27:1808 strongly excites
- L27:250 excites
- L27:1993 excites
- L26:1081 excites
- L26:2326 excites
- L27:243 weakly suppresses
- L27:781 weakly excites
- L27:957 weakly excites
- L27:1040 weakly excites
- L26:2423 weakly excites
- L27:2371 weakly excites
- L27:1669 weakly excites
- L26:502 weakly excites
- L27:1355 weakly excites
- L26:1125 weakly excites
- L26:1425 weakly excites
- L26:1043 weakly excites
- L26:934 weakly excites
- L27:1653 weakly excites
- L26:2402 weakly excites
- L27:824 weakly excites
- L26:1327 weakly suppresses opening parenthesis before an acronym
- L27:2851 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
- (ms
- (ns
- (ts
- fs
- (xs
- NS
- MS
- ns
- ns
- LS
Suppresses
- Oakland
- WARRANTY
- Hobby
- liệt
- Corbyn
- 髁
- ooke
- olland
- ạch
- radient