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. 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 an opening bracket that comes right after a full name, just before its short form, like "(NAMS)" or "(BAS)". Its strongest negative values fall on odd letters inside Finnish or other non-English words, so the pattern is mixed. Steering did not show a clear effect.
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.
-
last child, certain measurements may be linked with her projected lifespan, according to a study published Wednesday in Menopause, the journal of The North American Menopause Society (NAMS).⏎More specifically, leukocyte telomere length – telomeres
-
In early December, the East African Paediatric Association (EAPA) and Kenya Paediatric Assocation (KPA) in collaboration
-
another mother⏎The concept of digital twins, particularly when combined with MSPC, is relatively new, yet a well-constructed system based on Process Analytical Technology (PAT) will leverage both of these concepts, and as a result maximize your process
-
3⏎The British Safety Council International Safety Award 2011⏎Halley is the most southerly science research station operated by the British Antarctic Survey (BAS) and is located on the 150-metre thick floating
-
thread’s major diameter, minor diameter, pitch diameter, pitch (in case of metric threads) or threads per Inch (TPI in case of Standard American Equivalent (SAE) threads), flank angle, etc. Normally, a GO / NO
-
Policy, The Defence Advanced Research Projects Agency is developing some extremely high technology to tackle possible alien invasions, an example being the Extreme Accuracy Tasked Ordnance (EXACTO), a self-guided bullet, which aims to counter aliens moving
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 I don't depend on my first choice, I can choose my second choice. If I choose my second choice first, then I can choose my third choice, and so on, my choice depends on my first choice. I donap
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
.⏎Metsäpohjainen piha paransi päiväkotilasten immuunijärjestelmää jo kuukaudessa<|endoftext|>Research Looks at Alternative Power for Military⏎Source
-
linkit|⏎|Tila||Julkaistu - 1 syysk. 2021|⏎|OKM-julkaisutyyppi||A1 Julkaistu artikkeli, soviteltu
-
adopted the Genocide Convention. It defined genocide as “acts committed with intent to destroy,<|endoftext|>1905 - 1938⏎György Nemes was a graphic artist, active during the 192
-
been working as a graphic designer. He designed posters, brochures, album covers, etc. He many times cooperated with other artists, for example György Rad
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
- L4:2339 strongly excites
- L2:1699 strongly suppresses
- L18:249 strongly suppresses
- L5:779 strongly excites
- L19:390 strongly excites
- L19:287 excites
- L16:56 suppresses
- L2:1749 excites
- L18:221 excites
- L16:143 excites
- L0:49 suppresses
- L17:542 suppresses
- L1:20 excites
- L19:798 excites
- L1:141 suppresses
- L0:54 excites
- L17:886 excites
- L1:542 excites
- L6:401 suppresses
- L2:387 excites
- L10:2401 excites
- L18:924 weakly excites
- L1:56 weakly suppresses
- L16:17 weakly suppresses
Feeds later layers
- L27:1563 suppresses
- L24:2225 excites
- L24:725 excites
- L26:1203 weakly excites
- L26:789 weakly excites
- L25:1808 weakly excites
- L27:203 weakly suppresses
- L26:2111 weakly excites
- L27:2409 weakly excites
- L27:980 weakly excites
- L25:59 weakly excites
- L25:2710 weakly suppresses
- L24:156 weakly excites
- L27:1569 weakly excites first pieces of split words like clo and Le
- L27:2008 weakly excites
- L25:790 weakly suppresses
- L24:677 weakly excites
- L25:2358 weakly excites
- L24:90 weakly excites
- L24:94 weakly suppresses
- L25:194 weakly suppresses
- L27:1833 weakly excites
- L25:1564 weakly excites word pieces before 'trans' or 'transfer'
- L27:2136 weakly excites capital M inside abbreviations
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
- 았다
- 았
- omat
- kah
- ól
- fang
- omas
- ัส
- ika
- mal
Suppresses
- ä
- ös
- ök
- äm
- öt
- öh
- än
- äll
- ön
- Latter