unclear: nutrition and digestion lists
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 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 pin down. Its strongest moments come in health and food writing, such as the words "esophagus", "and" in a list of fats, and "in" before food types, but these look like a loose mix rather than one clear idea. Turning it up produced nonsense words like "acrophile", so the steering result does not help explain it.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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 a 75 watt bulb) which will give you more light<|endoftext|>The stomach is a key part of the gastrointestinal (GI) tract, sitting between the esophagus and duodenum. Its functions are to mix food with stomach acid
-
.⏎The pancreas is known as the enzyme “powerhouse” for the entire body, as it is responsible for the enzymes needed to break down carbohydrates, proteins and fats.⏎You may have an idea, but there are many kinds of digestive enzymes
-
, palm and coconut oils, sugary foods and beverages, saturated fat – found mostly in foods that come from animals; trans fat (trans fatty acids) – found in
-
All beans, legumes, grains and pulses; all soups, stews and casseroles; custard and yoghurt; steamed veggies; basmati, white, brown and sushi rice; mieliepap,
-
quate levels of vitamin D can have serious health consequences.⏎As a fat-soluble vitamin -- others are A, E and K -- vitamin D can be stored in the tissues of your body. This is in sharp contrast to water-soluble
-
no animal source foods (ASFs). Vitamin B12 is found in red meat, seafood, dairy products and eggs, and in fortified packaged cereals in developed countries
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 per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per per
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, and the impact that it had in their lives after the recovery of democracy in 1983.⏎The Navy Mechanical School (Escuela de Mecánica de la Armada
-
to recovery⏎Five years ago, Mosul Museum grabbed international headlines when the Islamic State (IS) group released footage of militants defacing priceless artefacts with sledgeham
-
, there is the risk of a major collapse in agricultural production in poor countries. This will increase malnutrition already at elevated levels in these countries.⏎Sr. Bénédicte Kikwengi, SNDdeN works with her students
-
any decision.⏎Keeps You Well-informed⏎If you consider all the facts before making any decision, then you will
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
- L24:533 weakly suppresses
- L2:1302 weakly excites
- L16:478 weakly suppresses
- L20:111 weakly suppresses
- L18:242 weakly excites
- L18:1442 weakly suppresses unclear: word fragments and digits inside numbers
- L19:256 weakly excites
- L1:1654 weakly excites
- L0:1406 weakly suppresses
- L16:1040 weakly suppresses
- L16:48 weakly excites
- L20:836 weakly suppresses
- L24:2590 weakly excites
- L20:149 weakly excites
- L0:366 weakly excites
- L0:638 weakly excites
- L16:60 weakly excites
- L9:498 weakly suppresses
- L0:1370 weakly excites
- L16:30 weakly suppresses
- L19:901 weakly suppresses
- L20:1574 weakly excites
- L1:2421 weakly suppresses
- L23:1408 weakly suppresses
Feeds later layers
- L27:2772 weakly suppresses
- L27:2541 weakly excites
- L27:1326 weakly excites
- L27:394 weakly excites
- L27:1654 weakly excites
- L27:2302 weakly excites
- L27:2540 weakly excites
- L27:205 weakly suppresses
- L27:834 weakly suppresses
- L27:2166 weakly excites
- L27:61 weakly suppresses
- L27:1764 weakly suppresses
- L27:2181 weakly suppresses
- L27:392 weakly excites
- L27:1843 weakly excites
- L27:1682 weakly suppresses
- L27:2157 weakly excites
- L27:1321 weakly suppresses
- L27:2258 weakly excites
- L27:1413 weakly suppresses
- L27:1920 weakly suppresses
- L27:2011 weakly excites unclear: word pieces near 'tag' and Q&A breaks
- L27:1470 weakly excites
- L27:2959 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
- ac
- urious
- tit
- egral
- (ac
- ateral
- ainer
- uz
- uros
- lique
Suppresses
- ibern
- boom
- 羽毛
- смотреть
- 有助
- 泊
- 杪
- bars
- 빈
- oggles