space right before an expected number
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 the empty space just before a number, such as the gap after 'more than' or 'an estimated'. It seems to be getting ready for a figure to come next. Its negative side fires on the decimal point inside numbers like 0.001. Turning it up gave repetitive filler text, so the steering does not show this idea clearly.
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.
-
students. Astronomy Center is sponsored by the American Astronomical Society, the American Association of Physics Teachers, and the NSF-NSDL.⏎Science.gov searches over 60 databases and over 2200 selected websites from 15
-
in academic publishing, have you ever wondered how many academic journals are there in the world? Well, I do.⏎It is estimated that there are currently more than 30,000 academic journals, and the number continues to increase by
-
rates (~99%) with very low false-positive rates (~0.2%). In other words, these tests are about 99.0% sensitive and 99.8% specific.⏎With performance like that one might expect these to
-
encies. LibGuides: Commerce Research Guide: Where can find journal ...⏎Functioning as a research and publishing network, ScienceOpen offers open access to more than 28 million articles in all areas of science. Although you do need to register
-
-so-simple question leads to an even scarier follow-up: What happens to us?⏎Coral reefs are the richest ocean ecosystems we have — they support an estimated 25 percent of marine species. According to a 2014 report
-
vitamin D. Our sun beds produce UV light that is about 95% UVA and 5% UVB.⏎Note that no supplementation should be substituted for proper medical treatment
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 two, three, or any, of the first, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or, or
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
mm) – this is one thousandth of a meter. In scientific notation, 1 millimeter is 1 – E3 m, which means 0.001 x 1
-
A single heat pulse will switch the magnetic orientation of a nanometer-sized magnetic grain in just a few picoseconds (1 picosecond is 0.000000000001 seconds). As a
-
% should take the following relationship: 1.000×25 / 100 , or , 1.000×0.25 .⏎Giving us a result of 250, this would be
-
), the nickel ($0.05), the dime ($0.10), the quarter ($0.25), the half-dollar 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
- L12:2885 weakly suppresses
- L11:71 weakly suppresses
- L18:1391 weakly suppresses
- L17:2453 weakly excites
- L10:287 weakly suppresses
- L9:1394 weakly suppresses number theory terms like 'number' and 'Gauss'
- L17:1773 weakly excites
- L1:1852 weakly suppresses
- L12:706 weakly excites
- L7:2699 weakly suppresses
- L2:431 weakly excites
- L12:1986 weakly excites
- L14:2166 weakly suppresses
- L12:496 weakly excites
- L6:3006 weakly excites
- L0:1807 weakly excites
- L17:2217 weakly excites
- L13:2726 weakly excites
- L6:2290 weakly suppresses
- L13:2928 weakly suppresses
- L5:1079 weakly excites
- L10:1423 weakly suppresses
- L10:2820 weakly excites
- L15:2758 weakly suppresses
Feeds later layers
- L22:1638 suppresses
- L21:1954 suppresses
- L22:1494 suppresses
- L21:1159 weakly suppresses
- L22:2597 weakly excites
- L23:929 weakly suppresses
- L23:2924 weakly suppresses
- L20:675 weakly excites
- L20:2351 weakly excites
- L25:783 weakly excites
- L23:478 weakly suppresses
- L23:2850 weakly excites
- L23:1534 weakly excites
- L22:1210 weakly suppresses
- L20:1691 weakly excites
- L26:2351 weakly suppresses
- L24:1960 weakly suppresses
- L20:1337 weakly suppresses
- L20:1103 weakly excites
- L22:2003 weakly suppresses
- L25:856 weakly suppresses
- L24:1701 weakly excites
- L23:2551 weakly excites baseball and sports mentions
- L23:2592 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
- ViewItem
- isks
- 畎
- ਯ
- يت
- بك
- igrams
- makt
- дей
- level
Suppresses
- Labour
- labour
- getResource
- urement
- 羔
- ilot
- enco
- erson
- حاسب
- enerator