the 'A' that starts a word fragment
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 fires on a capital A that begins the second half of a longer word or name, like the A in Ahab, Aids or Ama Dablam. When it was turned up, the model's writing just got rambling and off-topic, so the steering does 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.
-
, after Bayek left Siwa and A<|endoftext|>The mushroom spawn is any constituent that has been injected with mycelium –
-
.” While many in these professions verify this pessimistic outlook, does the Bible also discourage serving in such positions?⏎Consider wicked rulers in Old Testament times such as King Ahab and Queen Jezebel, who instituted Ba
-
al worship and slaughtered God’s prophets. Or King Ahasverus, who ordered the mass genocide of the Jews. Nehemiah faced
-
and her newfound Roman ally Julius Caesar joined with the Order and protected Lucius Septimius, one of their remaining members. Released from service yet feeling betrayed, Aya and Bayek founded the Hidden Ones to resume the fight against the Order,
-
, said it was every child's right to have proper access to sex education as well as psychological and social support.⏎Speaking in Geneva at a meeting on children and Aids organised by the Committee on the Rights of the Child, he said: "
-
u or Everest region⏎The major mountains are the Mt.Everest, Mt.Lhotse, Cho Oyu, Nuptse, Pumori, Ama Dablam
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: what if notimbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbitbit
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
: a peaked cap.⏎- adj. Having a sickly appearance: You're looking a little peaked today.⏎from Wiktionary, Creative Commons Attribution/Share-Alike License⏎- adj. Having a peak or peaks.⏎- adj. Sick
-
, Sweden, and Finland and the Kola Peninsula of northwest Russia. It is largely within the Arctic Circle.⏎from Wiktionary, Creative Commons Attribution/Share-Alike License⏎- proper n. The northern parts of Scandinavia and Finland with
-
, Creative Commons Attribution/Share-Alike License⏎- n. use, practise, exercise.⏎- v. To
-
in underwater installations and conduct, assist in, or be a subject of scientific research. Also called oceanaut.⏎from Wiktionary, Creative Commons Attribution/Share-Alike License⏎- n. An underwater explorer.⏎from WordNet 3.
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
- L23:2833 strongly excites
- L22:457 excites
- L19:1083 excites
- L20:1273 suppresses
- L21:1072 suppresses
- L18:102 suppresses
- L16:880 suppresses
- L11:689 excites
- L4:137 excites
- L19:751 excites the article 'a' before a noun
- L7:520 weakly suppresses
- L4:4 weakly excites
- L16:76 weakly suppresses
- L17:188 weakly excites 'a' before a noun phrase
- L20:174 weakly excites
- L21:518 weakly excites
- L19:2088 weakly suppresses
- L2:1866 weakly suppresses
- L6:357 weakly excites
- L20:901 weakly suppresses
- L6:1393 weakly excites
- L18:2208 weakly excites
- L19:283 weakly excites
- L18:296 weakly excites
Feeds later layers
- L27:657 weakly suppresses
- L27:2491 weakly suppresses
- L27:1528 weakly suppresses
- L27:2352 weakly excites
- L27:2512 weakly excites
- L27:2731 weakly suppresses
- L27:2099 weakly suppresses
- L27:1871 weakly excites
- L27:1461 weakly excites
- L27:2273 weakly suppresses
- L27:828 weakly suppresses
- L27:2309 weakly excites
- L27:2524 weakly excites
- L27:1826 weakly excites
- L27:686 weakly excites
- L27:609 weakly suppresses
- L27:475 weakly suppresses
- L27:2802 weakly excites
- L27:2378 weakly suppresses
- L27:692 weakly suppresses 'Ch' starting a proper name
- L27:185 weakly excites
- L27:1996 weakly excites
- L27:2578 weakly suppresses
- L27:2434 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
- arges
- sheets
- Award
- ten
- itionally
- cription
- ty
- sing
- tes
- 请您
Suppresses
- uctions
- zhou
- bit
- uction
- yen
- vore
- ABB
- bps
- 万户
- quí