the word 'same' before a noun
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. 8 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 "same" when it is followed by a noun or phrase that points back to something already mentioned, such as "the same dish" or "the same period". When it was turned up, the model's text kept coming back to the word "same" and to comparisons between things.
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.
-
consumption, Newgent said. Even though slow cookers are typically on for hours at a time, they will burn less energy than a traditional oven to prepare the same dish, such as a roast.⏎Look for ways to lessen the amount of time
-
we read and on what we read” - whether a book or journal, e-reader, phone or computer."⏎Other research has found declines in student comprehension when the same material is read on-screen instead of in
-
3-year period studied and can expect to live 79.2 years. Oklahomans, however, gained only 0.3 years over the same period, to 75.4. (For a table listing all the
-
prior to the advent of OCT, other imaging technologies were using ultrasound waves or MRI for imaging.⏎However, swept source OCT follows a little bit different mechanism for the same process. Unlike sound waves or MRI techniques, OCT uses light waves for imaging.
-
, less steam is produced because of a larger mass of water. A wide fog pattern of water passing through the same gases will produce an enormous amount of steam because of the greater surface-to-mass
-
ophen daily (> 500 mg) were 93 percent more likely to develop hypertension than those not taking the drugs, while younger women taking the same dosage were 99 percent more likely.⏎For ibuprofen, older women
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. So, [insert your name], [insert your name] [insert your name] [insert your names] [insert your names] [insert your names] [
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Ratifying CEDAW committed us to ‘being a society that promotes policies, laws, organisations, structures and attitudes that ensure women are given the same rights as men’.⏎Sex Discrimination Act passed⏎Australia passed the Sex Discrimination Act
-
of literature on air pollution and its relationship to cardiovascular diseases. The resulting scientific statement, which appears in the current issue of Circulation: Journal of the American Heart Association, concludes that air pollution is a serious public health threat and contributes to the development
-
digit of the second number. Each partial result should be aligned with the corresponding digit. That means the last digit of the partial product must be in the same column as the digit of the second number. No product may begin with any additional zeros.
-
) / membrane wall thickness⏎where D is the diffusion coefficient for the membrane material and is expressed in L2 / t.⏎Note: In gas release applications, mass is released from the tubing on a molecular basis and immediately dissolves in the groundwater
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
- L7:1508 excites
- L4:2695 suppresses
- L6:1851 weakly excites
- L6:1454 weakly excites
- L3:1944 weakly suppresses
- L3:1942 weakly suppresses
- L7:1088 weakly suppresses
- L7:1199 weakly suppresses
- L4:2376 weakly excites
- L2:84 weakly excites
- L0:1369 weakly suppresses
- L0:2856 weakly suppresses
- L1:2389 weakly suppresses
- L1:591 weakly excites
- L6:2317 weakly excites
- L6:1461 weakly suppresses
- L3:1567 weakly excites
- L5:2317 weakly suppresses
- L6:1437 weakly suppresses
- L3:2372 weakly excites
- L7:2196 weakly excites
- L5:2659 weakly excites
- L6:323 weakly suppresses
- L6:2498 weakly suppresses
Feeds later layers
- L10:1152 excites
- L12:2717 excites
- L10:411 excites
- L13:2305 excites
- L11:1430 weakly excites
- L11:2712 weakly excites
- L12:2426 weakly excites
- L18:851 weakly excites
- L14:1302 weakly excites
- L18:2881 weakly excites
- L12:3057 weakly excites
- L16:755 weakly excites
- L15:1216 weakly excites
- L14:2783 weakly excites the word 'same' in 'the same'
- L10:2989 weakly excites
- L17:2038 weakly suppresses
- L9:2725 weakly excites
- L14:926 weakly suppresses
- L19:940 weakly excites
- L19:2233 weakly excites
- L15:2919 weakly excites
- L9:1950 weakly excites
- L12:2525 weakly excites
- L22:1510 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
- ё
- OUTH
- -pattern
- same
- Ot
- achen
- ழ
- enchmark
- unce
- orphism
Suppresses
- xFD
- ayar
- 就知道
- 资产负债
- 珙
- Seas
- ,默认
- 漫长的
- 非常多的
- gev