comma inside stacked adjectives
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. 7 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
The neuron fires on a comma that sits between two describing words before a noun, such as "a two-story, wooden frame building" or "the five-story, 250,000-square-foot building". The steered outputs did not reflect this pattern in any clear way.
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.
-
will activate all evacuation horns throughout your home. Even if you are on another level of your home, you will be alerted of smoke or fire. A battery operated, single unit, smoke alarms don’t offer this type of protection because the units are
-
GFAP), and neurofilament light chain protein (NfL) — can help predict when athletes should return from sports-related concussions, a multicenter, prospective diagnostic study was launched and led by the Advanced Research Core (ARC) of
-
the country dedicated exclusively to Native Americans. The five-story, 250,000-square-foot, curvilinear building is
-
, 1904.⏎By December 1904 construction for the new Heilwood Company store had begun. It would be a two-story, wooden frame building and included not only a general store, but also the town post
-
3 weeks, until the new one arrived. In 1995, Mr. Paine presented the flag to the museum. The 48-star, Sterling Bunting flag is all wool and measures 60 high by 1
-
28. Gabriel Johnson then built the courthouse; the record is not clear about whether Walcott's or Johnson's plans were used. Costs of the two-story, brick building came
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 idea of a single person or a group to take on a single role. What does that mean? It is about the diversity of the roles and the variety of the tasks. The example of the single person, the single role. The
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
puts this to use. For now, we’re grateful for the lesson is Royer oscillators and rewinding transformers. But if you’d rather hack an off-the-shelf arc lighter, there’s always this arc lighter pyrograph
-
The answer. Histones. DNA is wrapped around these proteins to form a complex called chromatin and allows the DNA to be packaged up and condensed into a smaller and smaller space.⏎In almost all eukaryotes, histone-based chromatin
-
making nursing assessments, performing examinations, counselling, teaching and treating minor and self‐limiting illness. A nurse educator teaches in clinical and educational settings,
-
as that of hot-dip galvanized sheet.⏎4. Single-sided and double-sided galvanized steel square plates. Single-sided galvanized steel sheet, that
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
- L5:595 excites
- L6:141 suppresses
- L5:585 excites
- L4:615 weakly suppresses
- L5:1851 weakly suppresses
- L3:2525 weakly excites
- L5:2597 weakly suppresses
- L5:2885 weakly suppresses
- L6:289 weakly suppresses
- L1:2943 weakly suppresses
- L6:1556 weakly suppresses
- L6:12 weakly excites
- L5:1757 weakly excites
- L2:1003 weakly suppresses
- L3:779 weakly suppresses "but" after a "not just X" contrast
- L1:788 weakly suppresses
- L2:1584 weakly excites
- L6:479 weakly excites
- L3:444 weakly suppresses
- L5:782 weakly excites
- L6:1921 weakly excites
- L1:2196 weakly excites
- L6:199 weakly excites
- L5:2409 weakly suppresses
Feeds later layers
- L14:2458 excites
- L9:434 excites comma after an adjective in a list
- L12:2663 excites
- L17:2660 excites
- L8:386 excites
- L16:1715 excites
- L18:324 excites
- L9:2138 excites
- L12:1906 excites
- L13:976 excites
- L10:1481 excites
- L14:1337 excites
- L18:2110 excites
- L11:1174 excites
- L20:1976 excites
- L10:235 excites
- L15:2616 excites
- L9:120 excites
- L11:2733 excites
- L20:1388 excites
- L17:1280 weakly excites
- L9:392 weakly excites "and" inside organisation names
- L13:1365 weakly excites
- L11:630 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
- -controlled
- ce
- ManagedObject
- -low
- -initialized
- -run
- hollow
- strip
- Brooks
- .Observer
Suppresses
- .tk
- yard
- ím
- prite
- gatsby
- opyright
- .onNext
- plement
- pañ
- ffi