'of' after 'full'
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. 1 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 'of' when it comes right after 'full', as in 'full of clay' or 'packed full of'. Turning it up did not produce text about fullness, so the link shows only in the examples.
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.
-
associations may be buffered by the actual experience of working with ethnic minority students.<|endoftext|>As a gardener facing a yard full of clay, you don't have the luxury of planting without soil preparation. Increase your
-
unhappy when you risk losing them. This brain quirk has origins in survival strategies: Imagine you're a hunter-gatherer and you stumble on a pond full of delicious fish. Your dopamine soars and it feels great. You run back to
-
to the compost heap⏎Many people overlook the value of dirty washing up water as a compostable commodity. It is very nutrient rich water, full of small food particles and grease. It will help your compost heap degrade quickly, the
-
to mind. These are the common recyclable. But of late a new item – cigarette butt – has been added to the list. Used cigarette filters are full of toxins, which can leach into the ground and waterways, damaging living organisms
-
chip to replace entire rooms full of computers (Fig. 1).⏎The first microprocessor was the Intel 4
-
save your face and your eyes from the harmful rays of the sun.⏎Use swimming goggles⏎The water in swimming pools is full of chlorine, which protects you from germs and infections, but can actually hurt your
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 have been thinking about lately, and I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I am not sure if I have been thinking about lately,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and lateral; bracts and bracteoles absent; rays numerous, upwards-spreading, unequal. Calyx teeth obsolete. Petals yellow, obovate, mid rib conspicuous, apex with narrowly inflexed lobule. Sty
-
sequence mutations to arise in DNA from the mother or father. The exact mechanism for genome destabilization and how it’s inherited is unknown.⏎Lead author Yuri Dubrova is also cautious about how this data translates to humans. Since most cancer patients are
-
way to take advantage of the nutrients that milk provides.⏎The big breakthrough for yogurt came in 1905, when Bulgarian microbiologist Stamen Grigorov discovered the bacteria strain (lactobacillus bulgaricus) that fer
-
. But the real number could be ten times higher. The task of estimating the spread of HIV/AIDS in the Czech Republic falls to the Institute's Marie Bruckova, from the National AIDS Reference laboratory:⏎"I wouldn't say this is a huge
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
- L0:1301 suppresses
- L0:1224 suppresses
- L0:178 suppresses
- L0:2875 weakly suppresses
- L0:1459 weakly excites
- L0:2669 weakly excites
- L0:2661 weakly excites
- L0:44 weakly suppresses
- L0:2021 weakly excites
- L0:237 weakly suppresses
- L0:1195 weakly excites
- L0:102 weakly suppresses
- L0:407 weakly excites
- L0:2374 weakly excites
- L0:93 weakly suppresses
- L0:2715 weakly excites
- L0:2294 weakly excites
- L0:119 weakly suppresses
- L0:587 weakly excites
- L0:2414 weakly suppresses
- L0:157 weakly suppresses
- L0:107 weakly suppresses newline starting a new block
- L0:2148 weakly suppresses
- L0:1585 weakly suppresses
Feeds later layers
- L3:2094 suppresses "is" after "What" in questions
- L2:1238 suppresses
- L20:1494 weakly excites
- L10:2138 weakly excites
- L4:1108 weakly excites
- L12:1135 weakly suppresses
- L8:2339 weakly suppresses
- L13:1360 weakly excites
- L22:2623 weakly excites
- L6:2731 weakly suppresses
- L13:2832 weakly excites "by" after a passive verb
- L3:2925 weakly excites
- L19:1285 weakly excites
- L13:2881 weakly excites
- L3:2502 weakly excites
- L6:518 weakly excites
- L5:2262 weakly suppresses
- L15:18 weakly excites
- L16:816 weakly excites
- L13:2617 weakly excites
- L9:1733 weakly excites
- L17:1279 weakly excites
- L6:1542 weakly suppresses
- L16:1235 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
- alan
- itional
- ảng
- 可以获得
- igrams
- erd
- .functions
- 不甘
- Hon
- ="?
Suppresses
- ColumnType
- peri
- тика
- Same
- SEP
- ',)⏎
- ᴗ
- 下次
- لى
- BO