unclear: repeated or split word pieces
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. 16 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 is hard to read. Its strongest moments land on a word fragment or a repeated word, like the "qu" in "askutasquash" or the second "practice". Its most negative moments are on words like "that" and "summer". Turning it up or down did not show a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
10,000 years to Mexico and America. The name squash, which means uncooked or eaten raw, comes from the Native American word askutasquash.⏎Butternut squash tastes fairly mild, somewhat sweet, and somewhat nut
-
so that each can hear what the other is really trying to express and not just what they expect or want to hear.⏎Listening is a skill that takes practice, practice, practice! The good news is that we can get better at it.⏎What
-
outlines why labels are so important, and explains how to read them. It also provides some examples of appropriate medicated feed labels.<|endoftext|>Sentences with Failure, Failure in a Sentence in English, Sentences For Failure⏎1. The business was
-
1037th UGM doctoral graduate.⏎In conclusion of her dissertation entitled "Biosystematics of Murricata L. Annona, Annona squamosa and Annona Reticulata by Numerical Approach
-
more of it! But in a few days, dopamine spikes no longer came in response to the grape: The brain doesn’t waste dopamine on the same-old, same-old, but saves it for new information about rewards.⏎
-
Causal clauses are those that tell the reason why the main clause happened or was done. In Latin, these are introduced by quod, quia, and quoniam (among others). This video discusses when the indicative mood is used in
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 don't know of any. I don't know. I don't think I can't. I don't know. Maybe I should not be sure. I don't think of it as a question. I don't know. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
later hold a pencil for writing!⏎Why are art activities important?⏎Dr. Montessori once said, “Our schools prepare an eye that sees, a hand that obeys, and a soul that feels.” Through art activities, children learn how
-
Holes and gaps in the wall, roof, foundation, doors and windows allow air loss, which results in winter heat loss and summer heat gain. Tight construction, good insulation and high-performance windows are key. "
-
two classes, default on loan and no default on loan. There can also be multi-class classification which for example, is classifying
-
:⏎Time with you. By keeping your bedtime routine alive, you and your child get to do something new together — cheer for the good guys and boo the bad ones in the books you read. You also get a peek into how your
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
- L15:1443 excites
- L14:1631 excites
- L15:3058 suppresses
- L15:9 excites
- L8:512 weakly suppresses
- L14:982 weakly excites
- L14:2152 weakly suppresses
- L15:7 weakly excites
- L14:1010 weakly excites
- L15:127 weakly excites
- L14:2851 weakly suppresses
- L6:1203 weakly excites
- L14:748 weakly excites
- L13:2053 weakly suppresses
- L13:813 weakly excites
- L10:1126 weakly suppresses
- L10:641 weakly excites
- L10:163 weakly suppresses
- L14:1033 weakly excites
- L14:669 weakly suppresses
- L3:541 weakly excites the word 'a' after play or places
- L13:2069 weakly excites
- L7:2155 weakly suppresses
- L15:1128 weakly excites
Feeds later layers
- L17:327 suppresses
- L17:1002 excites
- L17:220 suppresses
- L20:2552 excites
- L17:1488 excites
- L19:211 suppresses
- L18:2520 suppresses
- L19:1037 suppresses
- L20:1910 excites
- L17:1567 excites
- L18:959 excites
- L17:459 suppresses
- L19:1019 weakly excites
- L18:1083 weakly excites
- L17:1096 weakly suppresses
- L18:468 weakly suppresses
- L19:1110 weakly suppresses
- L19:911 weakly excites
- L19:711 weakly excites
- L21:300 weakly suppresses
- L19:1563 weakly suppresses
- L19:431 weakly excites
- L19:2657 weakly excites
- L17:920 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
- “…
- 因为在
- ~~~~~~~~~~~~~~~~
- abruptly
- 派驻
- 要把
- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- teenth
- 将自己的
- 一直都
Suppresses
- oph
- avs
- 克
- den
- 易
- sno
- lo
- 佛
- fish
- Dog