'ago' after a span of years
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. 3 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 'ago' when it comes after a length of time, as in '66 million years ago'. Its negative side is on the word 'chain'. Turning it up did not bring out the time theme in the replies.
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.
-
66 million years ago - best-known for the extinction of the dinosaurs, the Cretaceous-Pale
-
families, educators, and other professionals.<|endoftext|>Tracking a Jurassic reversal of the Earth's magnetic field⏎Roughly 180 million years ago, during the height of the Jurassic period, the Earth's magnetic field flipped,
-
eylon-Tamils" who started coming from South-India over 1000 years ago and the "Plantation-Tamils" whose ancestors were brought by the British
-
has led to its popular moniker, the Cigar Galaxy. M82's burst of star formation was likely triggered a mere 100 million years ago in the latest of a series of bouts with another large galaxy, M81
-
quer table greeting us at the top of the⏎steps. Adorning each of the vessels are the origami butterflies, both⏎male and female.⏎Several years ago I⏎wrote a Shinto scholar who teaches at a major American university.
-
moved. The speed record is held by India, which started to make its way from East Gondwana to Eurasia about 140 million years ago at a speed of 20 centimeters per year.⏎Drilling into the
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 lack of a proper government of fire, the lack of a proper chain of trust, and the financial misfeasance of control, which is the chain of events, and so on. It's a, a Japanese company, and a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
endless chain is run, forming two hanging loops. In one loop is placed a movable block, whose pulley is shaped to take the chain. The load is attached to the movable block and is raised by pulling on the
-
the potential for chemicals to leach from the plastic into the contents of the bottle. PET is a type of polymer, which means it is made up of long chains of molecules called monomers. These monomers can be linked together in different ways
-
should regain consciousness.''⏎The message says that doing this somehow relieves blood pressure and eases symptoms.⏎Like most medical advice dispensed in the form of a chain e-mail message, it has no scientific basis. In fact, following its advice
-
of two pulleys of different radii connected and rotating as one on a common axle. The pulleys have their circumferences grooved and spiked so that a chain will run in them without slipping. Over the pulleys an
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
- L1:1 suppresses
- L2:2732 weakly excites
- L2:75 weakly suppresses
- L0:1088 weakly suppresses
- L1:2801 weakly excites
- L0:1301 weakly suppresses
- L2:2328 weakly suppresses
- L2:3038 weakly excites
- L2:2962 weakly suppresses
- L2:1468 weakly excites "over" in "take over"
- L1:1857 weakly suppresses
- L0:178 weakly suppresses
- L2:1036 weakly excites
- L0:1224 weakly suppresses
- L2:2711 weakly suppresses
- L2:2583 weakly suppresses
- L2:2102 weakly suppresses
- L0:2337 weakly suppresses
- L2:1899 weakly excites
- L2:2049 weakly suppresses
- L1:951 weakly excites
- L2:493 weakly suppresses
- L0:1774 weakly suppresses
- L1:2467 weakly suppresses
Feeds later layers
- L11:701 suppresses
- L4:1720 suppresses
- L19:3004 weakly suppresses
- L22:2356 weakly suppresses
- L24:2181 weakly suppresses
- L9:2141 weakly suppresses
- L27:1976 weakly suppresses
- L15:640 weakly excites
- L5:1743 weakly suppresses
- L10:616 weakly excites
- L8:2531 weakly suppresses
- L10:1343 weakly suppresses
- L12:824 weakly suppresses
- L9:690 weakly suppresses
- L12:2365 weakly suppresses
- L26:1578 weakly suppresses
- L7:1596 weakly excites
- L5:783 weakly suppresses
- L15:1427 weakly suppresses
- L12:1951 weakly excites
- L24:1542 weakly suppresses
- L12:1147 weakly suppresses
- L5:793 weakly excites
- L11:2873 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
- ated
- HEMA
- olum
- ician
- -plane
- ATED
- phia
- lectual
- сад
- elaide
Suppresses
- Lorem
- 长长的
- producción
- production
- baj
- production
- produção
- ök
- ��
- Production