the D in 3D or 2D
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. 20 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 letter D when it follows a number to mean dimensions, as in 3D printing or 2D graphics. When it was turned up, the written text drifted into unrelated topics and did not clearly talk about shapes or dimensions.
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.
-
, and the public congregations and usual places of divine worship; be<|endoftext|>New anti-flu drugs could be on the market within years after scientists create 3D map of part of the virus
-
time for thinking and ask that everyone be silent for a moment. Walk near that person. Empower others to confront.<|endoftext|>QR Codes are two dimensional (2D) graphic
-
new ones in the time required.⏎In response, Cristian Fracassi, CEO of Isinnova, a Brescia-based engineering firm, used 3D printing to meet the hospital’s demands and, in the process, save some of
-
the energy to physically check for yourself? Well, now you’re in luck: the University of Washington (UW) has invented a process in which 3D-printed objects connect to your WiFi – without the aid of conventional electronics – in
-
this emerging technology. The cost implications of this technology for healthcare facilities are considered, as well as the entrepreneurial opportunities that arise from the emergence and evolution of 3D printing.⏎Not long ago, it was thought that space travel was beyond the scope
-
plants has barely gone up. So this is now a public health crisis.⏎Surgeon and Professor of Urology⏎Over the past decade, 3D printing has garnered considerable attention due to its broad applications, its ease of customization,
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: the way a *poliagomor (poliagomor) is defined and how a *poliagomor is a *poliagomorum, and its subpart part. These are all parts of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ropes could be twisted together to form anchor cable.⏎Topics. This historical marker is listed in this topic list: Industry & Commerce.⏎Location. 38° 54.781′ N, 77° 3.
-
994323.html⏎"The Struggle for Power in the Ussr" StudyMode.com. 05 2012. 05 2012 <http://www.studymode.com
-
to actively build their nests. Ideally, the entrance hole should range from 1 1/8 to 1 1/2 inches in diameter.⏎- Entrance
-
1400 CE - 1800 CE⏎Oil on parchment attached to a limewood panel⏎|Dimensions||H: 11 1/16 in. (28.2 cm.), W:
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
- L18:649 strongly excites
- L16:598 suppresses
- L16:471 suppresses blank spot before a year
- L18:122 excites
- L18:881 excites
- L2:239 suppresses
- L19:244 suppresses
- L17:157 excites
- L6:1699 suppresses
- L0:258 excites
- L18:1939 excites
- L1:831 suppresses
- L19:2575 excites
- L1:1256 excites
- L16:19 suppresses
- L4:2532 excites surname pieces split across tokens
- L18:36 suppresses
- L15:12 excites
- L8:20 weakly suppresses
- L17:128 weakly excites
- L18:22 weakly excites
- L16:35 weakly excites 'not' in 'but not limited to'
- L16:1901 weakly suppresses
- L2:1417 weakly suppresses
Feeds later layers
- L21:339 weakly excites spot just before a year
- L21:110 weakly excites
- L21:1522 weakly excites
- L22:2389 weakly suppresses
- L22:318 weakly suppresses
- L22:851 weakly excites
- L23:2586 weakly suppresses
- L24:3034 weakly excites
- L21:613 weakly excites
- L27:2616 weakly suppresses
- L23:199 weakly excites
- L22:2879 weakly excites
- L23:785 weakly excites
- L24:2463 weakly excites
- L25:2756 weakly suppresses
- L24:209 weakly suppresses
- L22:951 weakly excites 'there' before are or is
- L23:1106 weakly excites
- L25:1890 weakly excites
- L21:1799 weakly excites
- L21:676 weakly excites
- L23:1323 weakly suppresses
- L25:2879 weakly excites
- L26:1602 weakly suppresses
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
- หว
- SITE
- erty
- S
- lesh
- เสร
- lesi
- icted
- SRC
- caps
Suppresses
- omen
- ereum
- 舅舅
- انيا
- anine
- aval
- cales
- urban
- rections
- umu