commas after a 'since' clause
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. 13 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 comma that ends a 'since...' or similar reason clause, where the main point is about to follow. Turning it up made the model drift toward talking about reasons and ideas, while turning it down gave less clear results.
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.
-
even at the expense of personal well-being. It can, however, be difficult to separate MIT’s culture from its people. Since the people help create the culture, one cannot ignore the fact that MIT’s students tend to be internally driven and place
-
the Fed injects more dollars into the economy it will show up as increased final demand/consumption and boost GDP. Since printing more money doesn’t create wealth, that doesn’t tell us much about our economic health (e.g., spending bo
-
what comprises their skeletons. Calcarea sponges have limestone skeletons. Hexactinellida have skeletons made of silica. (Since glass is made from silica, these sponges are nicknamed "glass sponges." ) Scleros
-
’s new queen likely settled on the jet’s nozzle to rest (hives often migrate to new locations with new queens) and since the hive never leaves its queen, everyone decided to have a kick-back right on the nose of the military’s sh
-
automation, robotics and many more. You can simply browse in the menu and on the page or leave a comment. Since new Raspberry Pi instructions and tutorials appear regularly, it is definitely worth activating the newsletter.⏎WiringPi is a library that is
-
you do buy some transplants, keep the pots they come in and reuse them⏎Yes, you can also buy containers. But with all these free versions above, there really is no need to unless you want everything to look consistent<|endoftext|>Whether it
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 of the three: The first is a number, and it is a number in the space, and it is a number: The first is a number: The first is a number: The first is a number: The first is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to the crest of the ridge where the best huckleberries grew.⏎Some years we got lucky. Every bush would be loaded, and we could be choosy—sampling deep
-
The studio probably had little choice but to create a form letter.⏎I don’t think the form letter nature of the rejection is necessarily impersonal. It sticks to objective criticisms. Probably you can’t pin a rejection on a
-
overwintering, and diapause—an insect version of hibernation.”⏎For example, “Boll weevils were thought to feed on pollen only, but my work reveals otherwise,” he says.⏎Nectars Could Be Where It
-
too dangerous to fly. Her flight home to Vancouver had been scratched because it connected through Frankfurt, an aviation super-hub shuttered because of the overhanging haze. Hickson, who works for a private geothermal energy company, acknowledged the
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
- L12:815 excites
- L12:801 excites
- L8:191 suppresses
- L11:2171 suppresses
- L12:731 excites
- L10:1677 suppresses
- L7:84 excites
- L12:834 excites
- L12:935 excites
- L11:2719 suppresses
- L10:2981 suppresses
- L8:309 suppresses
- L10:2621 excites
- L12:1672 excites
- L10:766 suppresses
- L11:2305 weakly excites
- L10:422 weakly suppresses
- L12:2718 weakly suppresses
- L4:2526 weakly excites
- L12:1796 weakly suppresses
- L7:766 weakly excites
- L11:2733 weakly suppresses
- L2:2373 weakly excites
- L5:246 weakly suppresses
Feeds later layers
- L14:1740 excites
- L14:1433 excites
- L14:756 suppresses
- L18:2592 suppresses comma after an opening clause
- L16:1393 suppresses
- L14:168 suppresses
- L16:2528 suppresses
- L14:848 excites
- L16:1702 weakly suppresses
- L15:997 weakly excites
- L15:2551 weakly excites
- L24:1429 weakly excites
- L15:1189 weakly suppresses
- L14:2212 weakly suppresses
- L16:2386 weakly suppresses
- L15:1847 weakly excites
- L14:1704 weakly excites
- L24:612 weakly suppresses
- L14:749 weakly excites
- L21:1490 weakly excites
- L15:318 weakly suppresses
- L16:1838 weakly suppresses
- L15:2328 weakly excites
- L16:861 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
- ault
- amenti
- evice
- cq
- istine
- -ret
- ody
- ewed
- ave
- lica
Suppresses
- |
- 部分
- IN
- 的部分
- .parts
- day
- part
- |^
- parts
- выраж