"one of" before a group
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. 19 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 "of" in phrases like "one of sixty images" or "one of the fastest", where something is being singled out from a larger group. When turned up, the model kept writing "many" and "a few".
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.
-
Voyager mission⏎The pale blue dot image was taken on 14 February 1990. It was just one of sixty images taken that day that were designed to capture most of the planets of the
-
solar cells that soldiers can roll up like a mat, transport, and unroll in a new location to start generating electricity on the spot.⏎It’s only one of the technologies South Dakota State University is exploring as part of a $10 million
-
And yet, experts predict more than 1 million computing jobs will be created in the next 10 years, according to the website.⏎Deister is one of the teachers, professionals, students and administrators nationwide pushing to increase access to computer programming
-
ite hit atmosphere at 64,000 mph⏎A meteorite that exploded as a fireball over California's Sierra foothills last April was one of the fastest and rarest known to have hit Earth.⏎A minivan-sized carbon
-
a number of cancers and scientists around the globe are busy trying to harness the power of the immune system in the fight against cancer. The HPV vaccine is just one of the first; expect to hear more and more about the topic in the future.⏎
-
The pair laid three eggs this year in February. None of their eggs hatched last year, the DNR said. The "EagleCam" is one of the program's projects funded by Minnesot
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 first element of the first element, the first element of the first element, the first element of the first first element, the first first first, the first first first first, the first first first first first, etc. What is the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
likely to develop psychiatric symptoms later on, and vice versa. Treatment for sleep disorders in young children may limit the development of psychiatric disorders later on.⏎The researchers interviewed the parents of nearly 1,000 toddlers. The toddlers were part of
-
wondered if he really died because I never got to see and hold his body."⏎The disease hit the DRC in August 2018 and has claimed the lives of over 2,200 people, according to the World Health
-
bursts and cosmic explosions, but it turns out that the spacecraft has been involved in 10 asteroid-observing sessions as well.⏎Swift's scientists had the satellite watch a couple of patches of sky that YU55 was predicted to
-
00 non-words.⏎To check whether the pigeons were learning to distinguish words from non-words rather than merely memorising them, the researchers introduced words the birds had never seen before.⏎The pigeons correctly identified the new words as words
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
- L17:30 suppresses
- L10:1377 suppresses
- L16:792 excites "the" inside a prepositional phrase
- L9:24 excites "the" before a named organisation or title
- L11:1168 excites
- L9:662 suppresses
- L17:831 suppresses
- L4:52 excites
- L10:120 suppresses
- L16:419 excites
- L13:1234 suppresses
- L11:875 suppresses
- L4:73 suppresses
- L16:2248 suppresses
- L18:1547 suppresses
- L18:171 excites
- L18:602 suppresses
- L15:1611 weakly suppresses
- L12:435 weakly suppresses
- L6:61 weakly suppresses
- L10:1162 weakly excites
- L4:217 weakly suppresses
- L5:2953 weakly suppresses
- L15:1848 weakly suppresses
Feeds later layers
- L21:822 strongly excites
- L22:825 strongly excites 'for' before a span of time
- L27:810 strongly excites
- L23:538 excites
- L21:2090 excites
- L20:2136 excites
- L23:2230 excites
- L24:2713 excites 'a' before few or variety
- L20:2987 excites
- L20:1935 excites
- L27:932 excites
- L26:1784 excites
- L20:2316 excites
- L24:1848 excites
- L21:785 excites
- L23:96 suppresses
- L20:2745 excites
- L23:870 excites
- L22:1358 excites
- L24:826 excites
- L20:1788 excites
- L27:545 excites
- L22:926 weakly excites
- L26:1083 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
- many
- several
- few
- 许多
- 很多
- 多种
- 若干
- Several
- 几种
- many
Suppresses
- himself
- "#{
- 也不例外
- 他也
- الآخر
- ("#{
- NDEBUG
- 做到
- lick
- 否则