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. 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 open bracket that comes right after a full name, just before its short form, like the "(" in "Test of Economic Literacy (TEL)". Turning it up made the writing wander into repeated, broken phrases, so the steering tells us little about it.
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.
-
ologic formations has been developed by Dr. von Wegner, a world-renowned senior geochemist at a national laboratory. The method, Thermal Oxygenation Xenolith (
-
a unique opportunity to make frequent assessments of economic literacy in U.S. classrooms. In this survey, student responses to test items from the Test of Economic Literacy (TEL) are used to create economic concept and content area achievement benchmarks. These benchmarks
-
Propulsion Laboratory manages the 2001 Mars Odyssey mission for NASA's Office of Space Science, Washington, D.C. The Thermal Emission Imaging System (THEMIS) was developed by Arizona State University, Tempe, in collaboration
-
Wahi) your nerves could not have endured it. Therefore, an interval was given in order to afford you peace and tranquillity." This state was experien
-
trillion in the current fiscal year that ends next September or nearly 5% of gross domestic product.⏎It’ll be the first time the deficit has topped the $1 trillion mark since 2012. Deficits topped $1 trillion for
-
to support WPA.⏎The WPA protocol implements much of the IEEE 802.11i standard. Specifically, the Temporal Key Integrity Protocol (TKIP) was
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: for an in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ia, which might be more disabling that the tics. Even though this complication has not been described in persons with Tourette's syndrome, the use of haloperidol is limited to the most serious cases.⏎Other medications in this group such
-
video:<|endoftext|>Tourette's Syndrome (cont.)⏎IN THIS ARTICLE⏎Treatment of Tics⏎The most effective medication for the suppression of tics is haloperidol (Haldol), a
-
and comet Kohoutek, whose sweep around the sun at that time was a scientific bonus. Daily
-
pre-exposure prophylaxis (PrEP) did not experience any protection against HIV compared to those in the placebo arm
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
- L24:2959 strongly suppresses
- L24:2012 excites
- L23:245 suppresses
- L23:363 suppresses
- L23:2818 weakly suppresses
- L23:801 weakly excites
- L22:3005 weakly excites
- L7:1247 weakly excites
- L24:2858 weakly suppresses
- L14:107 weakly excites
- L16:2977 weakly excites 'which' after a preposition
- L22:234 weakly excites
- L10:1572 weakly excites
- L1:257 weakly suppresses
- L11:395 weakly suppresses
- L6:1819 weakly suppresses
- L13:2505 weakly excites
- L18:2421 weakly suppresses
- L16:1613 weakly excites
- L17:1173 weakly suppresses
- L8:2816 weakly excites
- L15:1049 weakly excites
- L4:1279 weakly suppresses
- L1:2713 weakly excites
Feeds later layers
- L26:977 suppresses
- L26:367 excites
- L27:844 weakly excites
- L26:1785 weakly excites
- L27:2457 weakly excites
- L27:686 weakly suppresses
- L27:2003 weakly suppresses
- L27:2303 weakly suppresses
- L27:2265 weakly suppresses
- L27:1571 weakly suppresses
- L27:420 weakly suppresses
- L27:174 weakly suppresses
- L27:611 weakly suppresses
- L27:2238 weakly suppresses
- L26:1005 weakly suppresses
- L26:2439 weakly suppresses
- L27:2221 weakly suppresses
- L27:1347 weakly suppresses
- L26:304 weakly excites
- L27:440 weakly suppresses
- L26:1789 weakly suppresses
- L27:2107 weakly suppresses
- L26:1474 weakly suppresses
- L27:934 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
- trap
- tra
- Tracy
- Tr
- Tristan
- Tr
- tf
- .tim
- Trades
- Travis
Suppresses
- APTER
- ィ
- umph
- activated
- 운
- 颔
- yyyy
- -faced
- cks
- activation