'by' after 'known' or 'referred 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. 6 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 'by' when it comes after words like 'known', 'referred to' or 'called', as in 'known by the name'. Turning it up did not produce any clear, matching change in the written text.
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.
-
land at this season," said the boy.⏎We're afraid of the Ice Witch, who lives in Lapland, they answered.⏎You might know Samiland better by the name Lapland, home to the indigenous Sápmi people, numbering
-
redhorse — long known by the Cherokee and only recently recognized by western science — is an iconic species of the
-
11, 1858, created from the eastern half of the Minnesota Territory. The declare has a large number of lakes, and is known by the slogan the “Land of 10,000 Lakes”. Its
-
names and call them by name when participating in class. This helps to build a relationship of trust giving the
-
(PDPA) in 1965. Despite being regarded as an intelligent man, he was referred to as Najib-e Gaw (the Bull) by his opponents due to his physique. The PDPA staged a successful coup in
-
America with a mass market media campaign through the<|endoftext|>KAA Gent is often referred to by its second name: the Buffalos. At every match, you will hear
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 user is looking at the year 1, and the year is 12. This is a correct answer. The answer is: 120000 "00000" as the year is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
hours and half-hours⏎- I know that there are five- minute intervals between each number on the clock face⏎More Lessons for Grade 2⏎Common Core for Grade 2⏎Telling time on the analog and digital clocks (quarter
-
burg's Newberry Medal Winner.⏎This unit contains vocabulary, comprehension and writing activities for the book, The Magic School Bus: a Hurricane by Joanna Cole. Common Core RL.4.1, RL. 4.4, W 4
-
to help Grade 2 students learn to tell and write time from analog and digital clocks to the nearest five minutes, using a.m. and p.m.⏎Common Core: 2.MD.7⏎
-
- Cover pans: use only as much flame as necessary (Remember that slightly boiling water cooks foods at the same temperature as water that is rapidly boiling).⏎- Pressure cooker: use a pressure cooker more often since it cooks food fast over low heat.⏎
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
- L5:1393 suppresses
- L5:1857 excites
- L4:3042 weakly suppresses
- L5:584 weakly excites
- L4:97 weakly suppresses
- L3:1700 weakly excites
- L5:2882 weakly excites
- L5:180 weakly excites
- L4:580 weakly excites
- L3:230 weakly suppresses
- L0:705 weakly excites 'of' after 'pair' or a landmark noun
- L3:2140 weakly suppresses
- L4:2929 weakly excites
- L4:37 weakly suppresses
- L4:442 weakly excites
- L0:2599 weakly suppresses
- L4:1522 weakly suppresses
- L5:2482 weakly excites
- L3:2537 weakly suppresses
- L5:50 weakly excites
- L5:657 weakly suppresses
- L2:1524 weakly suppresses
- L4:21 weakly excites
- L0:2757 weakly suppresses
Feeds later layers
- L7:1635 weakly excites
- L27:516 weakly suppresses
- L27:1676 weakly suppresses
- L20:81 weakly suppresses
- L10:3021 weakly suppresses
- L16:2205 weakly excites
- L19:1652 weakly suppresses
- L22:2919 weakly suppresses
- L26:1358 weakly excites
- L25:2033 weakly excites
- L26:1213 weakly suppresses
- L24:289 weakly excites
- L14:95 weakly suppresses
- L13:3036 weakly excites
- L15:282 weakly excites
- L16:1638 weakly suppresses
- L20:896 weakly suppresses mid-word fragments in capitalised titles
- L21:732 weakly suppresses
- L8:1240 weakly excites
- L19:1138 weakly suppresses
- L14:2288 weakly suppresses
- L15:761 weakly suppresses
- L8:1224 weakly excites
- L25:2406 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
- reh
- [train
- 训练
- ’
- 猿
- ‘
- NewLabel
- pez
- 原先
- 股份
Suppresses
- mexico
- hers
- shire
- "';⏎
- lays
- akis
- addChild
- .omg
- dad
- MITTED