the digit 0 inside numbers
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. 0 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 digit 0 when it appears inside a bigger number, such as the 0 in 1540, 110, 10,000 or a year like 1950. When it was turned up, the model's replies did not clearly change in any single topic.
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.
-
right words⏎- Shortness of Breath⏎- Metallic taste⏎- Watery eyes⏎- Sensitivity to light⏎The ongoing inflammation can affect virtually<|endoftext|>10 June 1540 – Thomas Cromwell’s Arrest⏎Posted By Claire
-
becoming an increasingly popular alternative to lawsuits. As a result,<|endoftext|>10 / The Natural World⏎|Artist / Origin||⏎Albrecht Altdor
-
was crowned as the first king of Jerusalem on Christmas Day, 1100, by Dagobert of Pisa, the Patriarch, who would
-
ced later in his life, “and I often tremble when I ask myself the question: ‘What if I had not got the job?’”<|endoftext|>10 Tips to Boost Your Immunity⏎When it comes
-
used for seismic studies and six Lunar Module Ascent Stages that were crashed at the end of their missions, as was the complete Lunar Module from Apollo 10 and an assortment of orbiter probes that ended up impacting the surface.⏎That is
-
product. Be the first to review this product!⏎Your orders support book donation projects⏎This really is first rate customer service.⏎Search and browse over 110,000 wildlife and science products⏎Multi-currency. Secure worldwide shipping
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 have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
anchor cable.⏎Topics. This historical marker is listed in this topic list: Industry & Commerce.⏎Location. 38° 54.781′ N, 77° 3.647′ W. Marker
-
“weekends” are not recognised.⏎|Sunday||February 05||Kashmir Day||Observed since 1990 as
-
the Gates Foundation’s Measures of Effective Teaching (MET) project outlines the most important findings on research-based evaluation systems.⏎Teacher Evaluation 2.0⏎October 05, 2010⏎Teacher evaluations need serious fixing. Here’s
-
28||Eid-ul-Fitr Holiday||The end of Ramadan|⏎|Monday||August 14||Independence Day|⏎|Friday||September 01||Eid ul-Azha Day 1||Feast of Sacr
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
None: this is the first layer.
Feeds later layers
- L18:880 excites
- L4:667 excites
- L19:2848 excites
- L21:1031 excites
- L26:1617 weakly excites
- L18:137 weakly excites 'the' after 'to' in fixed phrases
- L13:1117 weakly excites
- L21:2399 weakly suppresses
- L10:273 weakly excites
- L4:320 weakly excites
- L6:30 weakly suppresses
- L3:1504 weakly excites
- L4:2938 weakly suppresses
- L26:389 weakly suppresses
- L20:164 weakly excites
- L16:3037 weakly suppresses
- L15:2638 weakly suppresses
- L22:2618 weakly suppresses
- L26:2106 weakly excites
- L10:1676 weakly suppresses
- L3:282 weakly suppresses
- L2:1523 weakly suppresses
- L6:2465 weakly suppresses
- L2:1624 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
- most
- anter
- 关键字
- 层
- dsp
- most
- iter
- مرا
- 邻里
- grandchildren
Suppresses
- roit
- arro
- rze
- ąż
- POSIT
- abol
- allas
- rices
- Tes
- htub