the 'pre' in pre-words
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. 26 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 start 'pre' of words like prehistoric, prehistory and pre-Civil War, where the word is split and 'pre' comes first. When it was turned up, the text drifted toward unrelated fragments and repeated phrases, so the push did not produce a clear theme.
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.
-
States.⏎Kahlo’s artistic sensibility and signature style developed in full. Her main source of inspiration was Mexican folk culture, which she blended with elements from pre-Colombian and Catholic mythology in her paintings.<|endoftext|>|Wednesday, 3
-
apparent spatial dimension to the site, it appeared possible that activity specific areas might be detected. Other reasons for selecting this site included the absence of knowledge concerning the late prehistory of this area, the rapid rate of erosion and destruction occuring
-
2011 - Nightlife during the Mesozoic era was full of activity and fraught with dangers, suggests a new study that found some dinosaurs and prehistoric reptiles were nocturnal, with carnivorous dinosaurs likely sneaking up on
-
while the old countries, populated since prehistoric times, each have their great groups of related motifs: the supernatural light creature
-
describes, “the piers of the bridge were festooned with statuary of mermen on dolphins. The bridge balustrades were ornately molded in pre-cast faux stone and lighting was generously provided from electric globes on elebor
-
It’s a 50-episode story that follows the time-traveling adventures of a little boy and his trusty computer companion, as they explore the prehistoric world. Listen to the episode here.⏎Now that you’ve listened to the
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: a one a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to the very essence of how the Universe expresses itself.⏎He begins by stating that all tathagatas are without Buddha nature although they have arrived in the state of reality. A
-
works such as Prayers or Meditations in 1545, she became not only the first English Queen
-
. People are strongly advised to take special care if a water supply contains two VOCs, however, even if they are both below the maximum allowable concentration.⏎Bioworx has just come out<|endoftext|>Nikola Tesla Articles⏎Tesla, a
-
Nature Beyond Conditions and a commentary by Rev. David Xi-Ken Astor, Sensei⏎BUDDHA NATURE BEYOND CONDITIONS⏎“All tathagatas are without Buddha nature, but at the same time, previously they have fully
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
- L16:25 weakly excites
- L2:1529 weakly excites
- L1:2340 weakly excites
- L23:206 weakly excites
- L8:10 weakly suppresses
- L16:110 weakly suppresses
- L17:1597 weakly suppresses
- L21:463 weakly suppresses
- L20:197 weakly suppresses
- L5:139 weakly excites
- L8:1802 weakly excites
- L18:2052 weakly suppresses
- L16:71 weakly excites
- L17:162 weakly suppresses
- L16:36 weakly suppresses
- L20:728 weakly excites
- L4:731 weakly suppresses
- L18:45 weakly suppresses
- L6:2269 weakly suppresses
- L10:1155 weakly suppresses
- L19:2685 weakly excites
- L8:3 weakly suppresses
- L6:468 weakly excites
- L16:15 weakly suppresses
Feeds later layers
- L27:797 suppresses
- L27:709 weakly suppresses
- L27:663 weakly suppresses
- L27:231 weakly excites
- L27:2732 weakly suppresses
- L27:2917 weakly excites
- L27:2474 weakly excites
- L27:2819 weakly suppresses
- L27:2029 weakly suppresses
- L27:1754 weakly excites
- L27:1143 weakly excites
- L27:644 weakly excites
- L27:1435 weakly excites
- L27:1472 weakly excites
- L27:1296 weakly suppresses
- L27:2883 weakly excites
- L27:1716 weakly excites
- L27:2003 weakly suppresses
- L27:1737 weakly suppresses
- L27:755 weakly suppresses
- L27:2587 weakly suppresses
- L27:1037 weakly excites
- L27:567 weakly excites
- L27:2236 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
- erate
- 虮
- состояние
- 骊
- Lud
- 期待
- 啡
- 工业大学
- �
- atif
Suppresses
- 一段
- 一个小
- 一个
- a
- 一个人
- ache
- 一個
- 的一个
- 一个多
- 一份