joining word in a list
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. 10 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 a connecting word such as 'and' or 'including' that comes right after a list of names or items. When it was turned up, the written text drifted into unrelated topics, so the steering results do 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.
-
International Space Station (ISS), making it the largest science lesson streamed live for the world to watch on YouTube. A prestigious panel of scientists, astronauts and educators which include His Royal Highness Prince Sultan bin Salman bin Abdul-Aziz Al Saud,
-
- Identify points where team members make decisions⏎- Note whether all members of the team are involved.<|endoftext|>The acronym SWOT stands for strengths, weaknesses, opportunities and threats. Organizations of all sizes use SWOT analysis to assess the effectiveness of their
-
the same as the height of landing, how fast was the athlete going in the horizontal and vertical directions right before landing? 2. Relevant equations vx = vcos(thata) vy
-
data structures that post-quantum algorithms will likely require.⏎We're indebted to Erdem Alkim, Léo Ducas, Thomas Pöppelmann and Peter Schwabe, the researchers who developed “New Hope”,<|endoftext|>Despite the continued
-
% of IBS sufferers have been helped with a low FODMAP diet.⏎FODMAPs stands for Fermentable, Oligo-, Di- and Mono-saccharides And Polyols. These are a type of carbohydrates that
-
experts say.⏎Prevent Blindness America urges consumers to attend only authorized public fireworks displays conducted by licensed operators. Individuals should not buy, use or store fireworks, including sparklers,
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 need to be... what? I need to be... I need to be... what... what? what... what? Okay, so the user is looking for a lot here, I need to be... I need to be... I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
rescue and medical personnel; and U.S. Navy and Coast Guard ships are deployed to support launch contingencies and astronaut recovery. Additionally, the Navy provides a KC-130 tanker for helicopter air refueling, E-2C aircraft
-
wolves.⏎tAsAM ruditashabdena gavAM haMbhAraveNa cha |⏎vrajasyotthApanaM chakrurghoShavR^iddhAH samAgatAH ||
-
01), systematically split into 6 supergroups, lettered from A to F (Lo et al., 2002). Those in supergroups C and D have coevolved with their nematod
-
What is the difference between 3-axis and 5-axis?⏎Three-axis machining is processed by linear feed axes X, Y, and Z.⏎5 axis cnc machining consists of linear interpolation movement of feed axes X, Y, Z and
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
- L5:149 strongly suppresses
- L2:1661 strongly excites
- L1:11 strongly suppresses
- L7:342 strongly excites
- L0:1010 excites
- L4:969 excites the letter C in vitamin C
- L1:340 excites
- L0:2910 excites
- L9:48 suppresses
- L3:516 excites
- L5:143 excites
- L7:287 suppresses
- L7:674 suppresses
- L2:114 suppresses
- L0:2939 weakly suppresses
- L0:1996 weakly excites
- L4:334 weakly excites
- L0:1388 weakly excites
- L6:944 weakly excites
- L1:50 weakly excites
- L0:2231 weakly excites
- L5:1712 weakly excites
- L0:2303 weakly suppresses
- L4:1860 weakly excites
Feeds later layers
- L18:88 strongly excites
- L16:374 strongly excites
- L17:241 strongly excites
- L20:314 strongly excites
- L19:381 strongly excites word-start 'acc' fragments
- L21:320 strongly excites
- L18:440 strongly excites
- L19:563 strongly excites
- L18:541 excites
- L21:493 excites
- L18:936 excites
- L22:607 excites
- L22:1155 excites
- L20:1486 excites
- L17:1216 excites
- L21:979 excites
- L22:2365 excites
- L18:816 excites
- L21:590 excites
- L27:2597 excites
- L18:1493 excites
- L26:851 excites
- L23:2641 excites
- L23:2107 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
- esium
- oftware
- apult
- oufl
- financed
- .PerformLayout
- 社
- tratt
- 郃
- Yugosl
Suppresses
- しい
- ower
- asted
- raham
- ixed
- emploi
- teacher
- ża
- enticate
- 迁