'said' in 'is said to'
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 'said' in phrases like 'is said to be', where a claim is passed on as a reputation. Turning it up did not produce clear output that matches this pattern.
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.
-
anan. There is a temple in the Tulunadu known by the name Yathugiri which is situated on a mountainous region. This temple is said to have 9 sacred water. This place is also known by the name Ya
-
politically or socially motivated purpose. The individual who performs an act of hacktivism is said to be a hacktivist.⏎Form Of Hacktivism :⏎Wikipedia is
-
risk factors also crop up as indicators in other diseases as well<|endoftext|>You are attending a highly exclusive seminar run by the famous artist known only as Leo.⏎Leo is said to be highly influential in the art world. In fact,
-
Mahadev Temple is said to be of ancient times. According to local legend, the temple was built by
-
which can then be glued to a different kind of wood to create a solid walnut appearance without the cost of actual solid walnut wood. Walnut heartwood in particular is said to be unique, as it appears to be deeply
-
that students have two important models of growth in the real world: linear and exponential. Have students compare and contrast them.⏎- The terms of an arithmetic sequence are said to grow “linearly.” The terms of a geometric sequence are said to grow
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 the relationship between the two things, the two things that are not only the two things that are not only the two things that are not only the two things. I have been thinking about the relationship between the two things
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
its employee, the bus driver.⏎When governmental immunity applies,<|endoftext|>Actinopteri (ray-finned fishes) > Cichliformes⏎(Cichlids, convict blennies) > Cichlidae⏎(Cich
-
brata (vertebrates) > Gnathostomata (jawed⏎vertebrates) > Teleostomi (teleost fish) > Osteichthyes (bony fish) > Class:⏎fish) > Stegocephalia
-
2: Measurement Adapter⏎Fig.3: Measurement Adapter Schematics⏎The Capacitors used are Mica (Glimmer) obtained from www.reichelt.de⏎With a jumper either the short or one of the series capacitors
-
large fish some 1.5 meters long. Its entire body is covered with scales, reminiscent of armor plating. It belongs to the Osteoichthyes class of bony fishes, of
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 weakly suppresses
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L0:1021 weakly excites
- L0:1360 weakly excites
- L0:451 weakly excites
- L0:732 weakly excites
- L0:120 weakly excites
- L0:84 weakly suppresses
- L0:143 weakly suppresses
- L0:851 weakly suppresses
- L0:30 weakly suppresses
- L0:2304 weakly excites
- L0:3070 weakly excites
- L0:1245 weakly excites
- L0:866 weakly suppresses
- L0:2278 weakly excites
- L0:116 weakly suppresses
- L0:1859 weakly suppresses
- L0:139 weakly excites
- L0:197 weakly suppresses
- L0:418 weakly suppresses
- L0:1831 weakly excites
- L0:1159 weakly excites
Feeds later layers
- L4:1118 weakly excites
- L21:2820 weakly excites
- L14:409 weakly excites
- L8:1432 weakly excites
- L6:738 weakly excites
- L2:247 weakly excites
- L7:2904 weakly excites
- L11:1781 weakly suppresses
- L19:925 weakly excites
- L4:1864 weakly excites
- L27:146 weakly suppresses
- L16:1008 weakly suppresses
- L7:213 weakly excites
- L5:2362 weakly suppresses
- L17:1551 weakly excites
- L2:2367 weakly suppresses
- L11:1645 weakly excites
- L18:2991 weakly suppresses
- L8:1812 weakly suppresses
- L9:1412 weakly excites the word "be" after modals and "to"
- L22:933 weakly suppresses
- L10:2744 weakly excites
- L10:605 weakly suppresses
- L5:1594 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
- oton
- ульт
- antly
- 烏
- нд
- nt
- äch
- ión
- 对此
- oreal
Suppresses
- amenti
- handleClose
- icina
- 廋
- лежа
- downtown
- atted
- 嫌疑
- Cooperation
- себе