narrator and speaker 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. 9 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 lights up on words that name the one telling a story or giving an opinion, such as "narrator" or "speaker", and on a few words like "sense" and "wonders". Turning it up made the model's replies drift toward everyday personal topics, though that link is weak.
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.
-
hand (the broken wall), and reminding us about the mysterious Something that destroys this wall.⏎- The Something is very different from the hunters and their dogs (the speaker knows all about them). The Something is far more covert. In fact, it
-
century. In fact, the only common thread was a mutual childhood memory of playing on a wreck of varying descriptions – depending on the age and recall power of the narrator.⏎In spite of the odds against finding the vessel’s identity, the public imagination
-
finally, that level of self-communion in which the husband will seek to explain his wife’s behavior and their crumbling marriage to himself.⏎Last, there is the narrator’s ambiguous silence, looking on dispassionately, allowing things as well as
-
as “…kingdom of dead…” inferno 8, line 85, wondering how this stranger has managed to penetrate their territory without being killed. The narrator refers to his leader as a master and mentor all through the play meaning his role
-
⏎Mary Oliver, a renowned American poet, has captivated readers with her profound exploration of nature and the human experience. Her poems often evoke a sense of wonder and contemplation, inviting readers to reflect on their own lives. In
-
ches of British-red.⏎One inquiring reader wants to know who controls the world's non-independent islands.⏎Detra K., from Los Gatos, California, wonders about 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: the number of times as many as the number of times as many as the number of times as many as the number of times as many as the number of times as many as the number of times as many as the number of times as many
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
://dictionary.reference.com/browse/daemon Pronounces it DEE-mon Which is how I've always said it. (The Golden Compass audiob<|endoftext|>50 Reasons to Eat & Grow Local Food⏎Preparing and preserving local food preserves culinary culture. For
-
for your middle school-aged kids. Teach them that you expect them to work hard. In return, you will reward them with plenty of downtime to do things they enjoy,<|endoftext|>
-
where we practised massage, yoga and relaxation.⏎We teamed up with P7 for an afternoon in Seaton Park where we played rounders and football. Thank you to our parent helpers who joined us and took part in the games.⏎Soup and
-
the world’s leading scientific expert on gratitude. a professor of psychology and founding editor-in-chief of The Journal of Positive Psychology, has discovered when people regularly engage in gratitude, they
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
- L6:2344 weakly suppresses
- L7:2429 weakly suppresses
- L4:2746 weakly suppresses
- L7:1426 weakly excites
- L3:1957 weakly excites
- L7:1106 weakly suppresses
- L8:1104 weakly excites
- L4:2335 weakly excites
- L8:425 weakly excites
- L3:3014 weakly excites
- L3:1630 weakly excites
- L6:6 weakly excites
- L8:1508 weakly excites
- L5:1320 weakly excites
- L8:1470 weakly suppresses
- L4:556 weakly suppresses
- L3:725 weakly excites
- L5:290 weakly suppresses
- L7:1051 weakly excites
- L7:1687 weakly excites
- L3:1259 weakly excites
- L7:63 weakly excites
- L8:4 weakly excites
- L1:2266 weakly excites
Feeds later layers
- L10:1910 weakly excites
- L11:431 weakly excites
- L12:2903 weakly excites
- L19:422 weakly excites
- L17:1337 weakly suppresses
- L11:3037 weakly excites
- L21:1095 weakly suppresses 'and' or ',' inside metric pairs
- L16:214 weakly suppresses
- L21:1861 weakly suppresses
- L13:2093 weakly excites
- L25:2141 weakly suppresses
- L18:2477 weakly suppresses
- L24:957 weakly suppresses
- L25:2595 weakly suppresses
- L19:2893 weakly suppresses
- L12:508 weakly suppresses
- L12:648 weakly suppresses
- L11:298 weakly excites
- L13:2687 weakly excites
- L13:1826 weakly suppresses
- L13:2989 weakly excites nouns for who or what a text is about
- L14:1748 weakly excites
- L11:2821 weakly suppresses
- L23:1406 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
- وع
- chop
- lean
- Comment
- 提议
- comment
- وص
- ETCH
- besie
- Comment
Suppresses
- 转载请
- 好吗
- став
- 么
- 全校
- Bins
- ğer
- État
- 太多了
- oggle