the word 'flat' as adjective
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 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 lights up on the word 'flat' when it describes a surface, area, or shape, like a flat surface or a flat spiral galaxy. When it was turned up, the text drifted toward surfaces and shapes, though not always in a clearly related way.
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.
-
are many kinds of digestive enzymes that break down different kinds of food items.<|endoftext|>Chinese architecture started from ancient cave-dwellings, that are still inhabited on the flat yellow plains in varioust parts of Northern China. You ca be sure not to
-
A sheep to be measured shall stand on a flat dry surface. Sheep will be shown in its natural fleece (not shorn).⏎
-
of Columbia, MO⏎Camera: SBIG ST8300 with FW5⏎Telescope: William Optics FLT110 w/Reducer Flattener IV system @ f5.
-
Election pertaining to Blank Electoral College Map 2016 Printable, Source Image : clanrobot.com⏎A map is identified as a counsel, usually with a flat area, of a complete or component of a region. The job of any map
-
Desert of Mauritania is easily visible from space because it is nearly 50 kilometers across. Once thought to be an impact crater, the Richat Structure's flat middle
-
the bonsai pot. If the soil is dry at this depth, the bonsai tree needs watering.⏎Take your bonsai tree outdoors and set it on a flat surface. Fill a watering can with water and tip the spout downward at a
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 first step in the development of a mixture of... the... is... is... is... is... is... is... is... is... is... is... is... is... is... is... is... is... is... is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
The rest of the hole would also receive new soil, not the clay. My theory was that perhaps the roots would grow down as well as out and make the tree stronger. My question is whether the extra digging and new soil was worth the effort
-
shallower in heavy clay soils.⏎In some situations, it can be detrimental to amend or replace heavy clay backfill soil, because it can trap water around the tree's roots and also can impede root development.⏎Water will tend to stay in
-
ain from fertilizing the shrub from early summer onward to prevent new growth that may be damaged by winter temperatures.⏎Refrain from pruning which can stress the plant and create entry points for disease over the winter.⏎Maintain cool ambient temperatures for
-
ground until the soil is warm (65 to 70 degrees F) and the weather has settled. Even a 50-degree night will slow the plant's growth for some time afterwards.⏎You can start basil seeds indoors, three to
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
- L3:1042 excites
- L3:464 weakly excites
- L2:49 weakly suppresses
- L3:102 weakly suppresses
- L1:364 weakly excites
- L2:1657 weakly excites
- L3:50 weakly excites
- L2:2664 weakly excites
- L2:1036 weakly excites
- L3:2544 weakly suppresses
- L3:691 weakly excites
- L2:2328 weakly suppresses
- L2:380 weakly suppresses
- L3:2725 weakly suppresses
- L3:1759 weakly suppresses
- L1:2677 weakly suppresses
- L3:115 weakly suppresses
- L2:506 weakly suppresses
- L1:1507 weakly suppresses
- L3:59 weakly excites
- L3:229 weakly excites
- L1:2801 weakly excites
- L1:212 weakly suppresses
- L2:2258 weakly excites
Feeds later layers
- L5:2812 excites
- L19:982 weakly suppresses
- L7:3043 weakly suppresses
- L7:2457 weakly excites
- L13:466 weakly suppresses
- L19:1893 weakly excites
- L21:2700 weakly excites
- L10:726 weakly excites
- L6:165 weakly excites
- L7:1813 weakly excites
- L6:166 weakly excites
- L24:707 weakly excites
- L11:660 weakly excites
- L7:1728 weakly suppresses
- L5:2891 weakly suppresses
- L10:1133 weakly excites
- L25:1627 weakly excites
- L12:1192 weakly excites
- L5:1632 weakly suppresses
- L13:2730 weakly excites
- L14:2493 weakly suppresses
- L7:756 weakly excites
- L7:1906 weakly suppresses
- L7:2390 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
- flatten
- wyjątk
- allis
- 平坦
- Można
- Dé
- 平整
- lokale
- кая
- exter
Suppresses
- ัย
- getParam
- ited
- Min
- ический
- مة
- choices
- ⼩
- .MinValue
- uchos