soil and loam
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. 4 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 reliably lights up on the words for soil and its types, especially 'loam', 'soil' and 'sand, clay and loam'. Turned up, the model's writing kept coming back to soil, though only some of the samples did so. Turned down, it tended to drift toward medical or genetic conditions instead.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
plants. Three basic soils are sand, clay and loam. Sand consists mainly of quartz particles, and clay is made up of extremely fine
-
how soil is formed.⏎- Recognize that life depends on soil.⏎*See glossary for definations⏎Beakers with water⏎Samples of sand, loam and clay⏎Seeds for planting⏎Toothpicks or forceps⏎
-
to our dependence upon soil.⏎- Ask participants to tell you what soil is made of. Invite them to examine the soil samples (sand, loam and clay) under the microscopes.⏎- Bring the class together to discuss what
-
Thoughts⏎What Is Soil?⏎The term “soil” comes from the Latin word solis, meaning “ground”. In English,
-
include information about the distribution of rock such as limestone, sandstone, coal and granite. Surficial material maps describe deposits that occur above bedrock. This includes soil, but also details up to several hundred feet of deeper unconsolidated material
-
⏎Australia - northern Western Australia, northern Northern Territory, northern Queensland⏎Flats in woodland or vine thicket and on river banks, growing in sand, loam, cracking clay, calcareous or granitic soils; at elevations up
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 value of a variable is the same as the value of a variable in the table of contents. The value of the variable is a single line. The value of the variable is a single line. The value of the variable is a single
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
1 in 7,000 births and the risk increases with maternal age. The conditions are associated with severe mental handicap and several physical defects. Most affected individuals die before or soon after birth and they rarely survive beyond the first year of
-
in the affected joint. This condition most commonly affects the joints in hips, knees, hands, and spine. Rarely, the
-
by mutations in the PTPN11 and RAF1 genes.⏎The symptoms and physical characteristics associated with LEOPARD syndrome are highly variable. However, most affected individuals tend to exhibit characteristic abnormalities of the skin, the structure and function of the
-
can cause severe turmoil in a person's physical, psychological, and social health. Eating disorders most commonly affect women and occur in adolescence and young adulthood, though they can occur in anyone,
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
- L2:2366 suppresses
- L3:1076 weakly suppresses
- L0:2073 weakly suppresses
- L3:342 weakly excites
- L3:2782 weakly excites
- L3:1029 weakly suppresses
- L3:272 weakly excites
- L1:1526 weakly suppresses
- L2:2408 weakly suppresses
- L3:2140 weakly suppresses
- L3:2763 weakly excites
- L3:1965 weakly excites
- L3:2938 weakly suppresses
- L2:2047 weakly excites
- L2:2604 weakly excites
- L2:49 weakly suppresses
- L1:2801 weakly excites
- L1:162 weakly excites
- L3:464 weakly excites
- L3:2077 weakly excites
- L0:1301 weakly suppresses
- L3:2202 weakly excites
- L2:1479 weakly excites
- L2:2175 weakly excites
Feeds later layers
- L6:1155 excites
- L10:2241 excites
- L7:1585 excites
- L6:2153 excites
- L5:1844 excites
- L10:3067 excites
- L10:1802 weakly excites
- L8:1028 weakly suppresses
- L5:2300 weakly excites
- L11:2813 weakly excites
- L6:1503 weakly excites
- L7:2927 weakly excites
- L5:2809 weakly excites
- L9:427 weakly excites
- L26:2710 weakly suppresses
- L17:1967 weakly suppresses
- L27:786 weakly excites
- L11:2454 weakly excites
- L5:2612 weakly excites
- L7:2152 weakly excites
- L7:1820 weakly excites
- L8:1447 weakly suppresses
- L21:1811 weakly suppresses
- L24:567 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
- erre
- 堆积
- itty
- ough
- uffy
- 以来
- youthful
- ffective
- rede
- 沙滩
Suppresses
- Vien
- iability
- uitar
- pir
- .btnClose
- ставил
- icone
- -vector
- icensing
- PLICATION