the word 'new'
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. 8 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 responds to the word 'new' in phrases like 'the new city', 'build new ones' and 'new blood vessels'. Turning it up did not produce any clear theme in the continuations, so the steering shows nothing about this word.
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.
-
to uphold Ma'at, the principle of order and truth. At death, the Pyramid Texts stated that the pharaoh was to become the sun or the new Osiris. However, the journey of this transformation was perilous. The Pyramid
-
specific period⏎of time; at the conclusion of this period, this world will necessarily end and⏎its earth and heavens shall be replaced by a new earth and new heavens; a new⏎world will then come into being. All mankind shall be
-
Sometimes it makes sense for a function to return multiple things. In such cases, it's common to just return them in a tuple: "return (count, new_obj)".⏎Sometimes you might want to return objects of different types based on whether
-
and deserve full citizenship. They are fighting for their rights and their place in the new scene after the revolution. However their struggle exemplifies the many of the issues that
-
28 grams of protein into your diet. This is 13 percent of the daily value for protein, a nutrient your body needs to repair cells and build new ones.⏎Source of Minerals⏎Tamales contain good amounts of iron and phosph
-
are carried over to the new data directory.⏎Stop the existing MySQL instance using the data directory. This must be
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 to be honest with you that I have no time to write for you what I can say. And I also have no time to write for you in the same way. I have been thinking about this the same way I have been
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of the earth covered in saltwater, many plants and animals live along, on, and under the ocean. Ocean habitat consists of three zones- the Intertidal shoreline zone, the Pelagic or open ocean zone, and the deep Abyssal
-
is generally agreed to be responsible for up to 85% of the visible signs of skin aging! The good news is that, in comparison to intrinsic skin aging, it is
-
the signs of skin aging. Now that you<|endoftext|>America's first industry was neither textiles, nor iron,
-
of sugary or acidic foods and beverages, and drink plenty of water to help wash away food particles and neutralize the acid in your mouth.⏎5. Floss on a daily basis:⏎Flossing removes food particles and plaque that can become
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
- L4:2828 excites
- L4:1906 excites
- L7:283 suppresses
- L3:1615 weakly suppresses
- L0:134 weakly suppresses
- L7:1753 weakly excites
- L2:641 weakly suppresses
- L4:2459 weakly excites
- L2:572 weakly suppresses
- L6:170 weakly suppresses
- L7:1521 weakly suppresses
- L4:1683 weakly suppresses
- L2:2903 weakly suppresses
- L5:1310 weakly suppresses
- L0:1893 weakly suppresses
- L7:2550 weakly excites
- L0:1106 weakly excites
- L7:2899 weakly suppresses
- L6:261 weakly suppresses
- L2:1447 weakly excites
- L7:700 weakly suppresses
- L4:2860 weakly suppresses
- L4:1931 weakly suppresses
- L5:698 weakly excites
Feeds later layers
- L11:1755 excites
- L19:1042 excites
- L15:1892 excites
- L11:1713 excites
- L14:2743 excites
- L11:1895 excites
- L16:2123 excites
- L21:2084 excites
- L9:409 weakly suppresses
- L12:1122 weakly excites
- L11:239 weakly suppresses
- L9:743 weakly excites
- L9:750 weakly excites
- L9:803 weakly excites
- L12:2917 weakly excites
- L16:887 weakly excites
- L12:480 weakly suppresses
- L17:1620 weakly excites
- L12:1854 weakly suppresses
- L14:832 weakly excites modifier words before their head word
- L13:71 weakly suppresses
- L9:1215 weakly excites
- L16:2395 weakly excites
- L9:1475 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
- replacements
- 替代
- skirts
- ования
- 取代
- replaces
- improves
- 重建
- versions
- improvements
Suppresses
- apot
- 帻
- apest
- apat
- nuit
- çon
- Chin
- 一杯
- flag
- acist