words leading into food, plant or habitat details
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. 21 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 lights up on the word just before a detail about what animals eat, where they live, or what plants and grass are like, such as 'made up of', 'diet of' or 'including'. When it was turned up, the writing drifted toward trees, leaves, grass and green plants.
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.
-
garden. And, it can be exciting to build your new landscape design around a mature native tree.⏎Native plants can attract native animals such as butterflies and birds by providing food and shelter. Fruits such as acorns or berries are vitally important
-
of fibre and roughage and needs to be available to your rabbit to munch throughout the day. 70% of your rabbit’s diet should be made up of hay
-
some notable behavioral patterns. Generally, deer feed in the late evening or the early morning. They are selective browsers with a variety of foods making up their diet, including broad-leaved herbaceous and woody plants, and grasses. They will
-
the same reason pygmy marmosets are threatened. Pygmy marmosets could be endangered easily. Tamarins have the same diet of sap, fruit, small insects and small buds.
-
rather than pasture, contributes to stress (and ulcers). So does any significant change in the horse’s routine or life style, such as confinement with little or no turnout or grazing, trailering, training, lay-up due to sickness or injury
-
consists of small fish and aquatic invertebrates.⏎The bird is monogamous and colonial. It builds a stick platform nest usually in reeds, but sometimes in trees over water.<|endoftext|>Anti-platelets are a group of medicines that stop
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: if you don't care about the process, you don't care about the story. How does this process or story relate to the | | | | | | | | | | | | | | | | | | | | | |
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Moisture: Provide ample water during establishment. Once established, irises need little care. Hardiness: USDA Zones 4-10. Propagation: The creeping underground stems (rhizomes
-
house doesn’t receive a lot of sunlight, so I purchased a grow light, and decided I would grow my plants inside. I learned that bean plants need about 4-6 inches of rain each week, so once every two days, I
-
ible blooming houseplant (or garden plant in USDA Zones 9 and warmer). Its needs differ from average houseplant care because it likes dry
-
a 6- to 8-week rest, move the plant into a warmer room and begin watering more frequently; blooming should soon commence.⏎USDA Zones 9
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
- L20:2646 excites
- L20:1254 excites
- L20:1722 suppresses
- L19:1474 suppresses
- L7:2669 suppresses
- L15:2975 weakly suppresses
- L20:1627 weakly excites
- L19:2102 weakly excites
- L11:1287 weakly suppresses
- L16:2963 weakly excites
- L9:1401 weakly excites
- L5:110 weakly suppresses
- L18:2756 weakly excites
- L16:1281 weakly excites
- L17:2448 weakly suppresses
- L1:3033 weakly excites
- L1:2172 weakly suppresses
- L14:2173 weakly suppresses
- L10:1436 weakly excites
- L7:1610 weakly suppresses
- L11:1451 weakly suppresses
- L20:611 weakly excites
- L18:2353 weakly excites
- L16:845 weakly suppresses
Feeds later layers
- L24:2635 excites
- L25:1282 weakly excites
- L22:2812 weakly excites
- L23:2637 weakly suppresses
- L22:938 weakly excites
- L25:2243 weakly suppresses
- L26:1114 weakly excites
- L22:2843 weakly excites
- L23:1082 weakly excites
- L22:1728 weakly excites unclear: code assignments and parentheses
- L22:633 weakly suppresses
- L24:450 weakly suppresses
- L25:2556 weakly excites
- L24:490 weakly excites
- L25:1627 weakly suppresses
- L25:2781 weakly excites
- L24:1815 weakly excites
- L25:2572 weakly excites
- L22:2576 weakly suppresses
- L25:1803 weakly excites
- L25:157 weakly excites
- L23:1806 weakly excites
- L23:3051 weakly suppresses
- L27:1023 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
- 树叶
- 树枝
- 草
- branches
- бер
- 榛
- grass
- 草地
- hay
- surfaces
Suppresses
- 不下
- udos
- illery
- peaker
- emo
- pron
- okens
- .testng
- 实施细则
- 嗝