hair and hair anatomy 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. 23 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 the word 'hair' and on nearby words in texts about hair, hair structure or plant parts. When turned up, the model started writing about roots and root systems.
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.
-
transparent, overlapping scales, it resembles the shingles on a roof. The cuticle's primary purpose is to safeguard the inner structures of the hair, such as the cortex and medulla, from damage caused by external factors like heat, chemicals,
-
for our landfill.⏎There are lots of products you can switch to:⏎- Instead of bottled water, change for tap water in a water bottle⏎- Swap your hair washing products for shampoo and conditioner bars⏎- Say goodbye to shower gel and hello
-
, WWI Map Is Up for Auction⏎- Thousands Of FBI Documents About Civil Rights Era Destroyed By Flooding⏎- Ancient Egyptian Woman with 70 Hair Extensions Discovered⏎- Europeans drawn from three ancient 'tribes'⏎- Conservatives
-
affected by electrical noise emitted by gadgets like smartphones, TVs, computers, even vacuum cleaners and hairdryers, AM signals are prone to distortion and crackling.⏎That crackle
-
and personal-care products (soaps, shampoos, hair sprays, and nail polishes).⏎Phthalates are used widely in poly
-
.⏎- What is the turkey's beard made of? Turkey beards are actually comprised of bristles or filaments that appear to be hair-like, modified feathers known as mesofiloplumes. The individual bristles emerge
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 can't find the map to the first two books in the book map and I need to find a way to move around within the book map. The first two books are important to me, so I need to understand the pattern of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the top to 5 lobes, the lobes longer than the tubes and becoming reflexed.⏎Fruits: Berrylike, globose, 5-6 mm across, shiny red or purplish-black, occasionally brown, yellow,
-
ile, style 1, entire or 3-branched⏎Fruit: capsule, loculicidal, thin walled, ± rounded, 6-angled, or winged, chambers 3⏎Seeds many,
-
Citrus reticulata) is a tree from the Rutaceae family producing citrus fruits called mandarins. The fruit is spherical and slightly flattened, with a diameter of 5 to 9 cm. Its flesh, sweet and frag
-
vate, mid rib conspicuous, apex with narrowly inflexed lobule. Stylopodium conic; styles very short, reflexed. Fruit oblong, terete, glabrous; ribs 5, acute or round-obt
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
- L20:1268 suppresses
- L21:2902 suppresses
- L9:3018 weakly excites
- L17:1534 weakly excites
- L7:204 weakly excites
- L20:2644 weakly excites
- L20:963 weakly suppresses
- L22:1122 weakly suppresses
- L19:1324 weakly suppresses
- L8:1612 weakly excites
- L11:2202 weakly suppresses
- L5:1267 weakly suppresses
- L2:483 weakly suppresses
- L10:852 weakly excites
- L19:2860 weakly suppresses
- L5:216 weakly suppresses
- L6:1413 weakly excites
- L2:1321 weakly suppresses
- L4:2285 weakly excites
- L0:2609 weakly suppresses
- L22:2106 weakly excites
- L18:2783 weakly suppresses
- L18:1460 weakly suppresses
- L22:254 weakly suppresses
Feeds later layers
- L25:2202 weakly excites
- L25:2725 weakly suppresses
- L25:2037 weakly excites
- L25:84 weakly excites
- L26:1543 weakly excites
- L24:2635 weakly excites
- L27:2534 weakly excites
- L25:1610 weakly excites
- L24:1947 weakly excites
- L27:1668 weakly excites
- L24:712 weakly excites 'long' in 'long-term' and 'long-lasting'
- L24:3013 weakly excites
- L24:2229 weakly suppresses
- L24:1418 weakly excites
- L24:1892 weakly suppresses
- L25:2894 weakly excites
- L24:69 weakly excites
- L26:2844 weakly excites
- L25:1290 weakly suppresses
- L26:2000 weakly excites
- L24:615 weakly excites
- L25:2950 weakly suppresses 'in' in 'in progress' phrases
- L25:2036 weakly excites
- L24:2280 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
- 根
- 茎
- 枝
- root
- -root
- Branch
- heads
- boom
- (axis
- 直
Suppresses
- .checkSelfPermission
- че
- Subaru
- atches
- atcher
- RI
- essel
- isto
- éric
- iore