unclear: sun angles, geometry and road-safety terms
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. 9 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 words about angles, slopes and directions, like 'perpendicular to' or 'low-angle sun', and it is quiet or pushed well below zero on text about visiting or travelling to places. When it was turned up, the model drifted into maths problems and word puzzles.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
long shadows produced by the low-angle sun to lead the viewer's eye into the photo.⏎Shadows extending toward the
-
See also single-crystal silicon and polycrystalline silicon.⏎Angle of Incidence — The angle that a ray of sun makes with a line perpendicular to the surface. For example, a surface that directly faces the sun has a solar
-
light is off.⏎Mirrors and Windshield⏎- Mirrors should be clean and adjusted properly from the inside.⏎- Wind
-
different designs will work better in different spaces. However, some basic tips include making sure the stairs are wide enough for comfortable use, and that they have a gradual slope rather than a steep incline<|endoftext|>Teaching in any Setting: Participation Project Environment
-
coming headlights, rather look toward the right side of the road and use the white fog line as a guide. When headlights from vehicles following you reflect in your rearview mirror, use the “day-night” feature on the mirror or adjust your mirror
-
ed from iCoachMath]⏎Focus:Focus of a parabola lies on the axis of symmetry.⏎Directrix:Directrix is a line that is perpendicular to the axis of symmetry of a parabola.⏎Geometric figure and Formulas
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: San Francisco, San Francisco, (4x3x) San Francisco – San Francisco on the way, which is the best way to get to the city. I have been staying in a 5-star hotel, which is the best way
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
The lessons take place on the school soccer field, where groups of students practice setting up tents, learn how to use a compass, are introduced to "Leave No Trace" ethics, and learn the basic skills needed to plan a trip or overnight stay
-
strong currents and winds make landing on the few steep rocky shores a treacherous task, even in the zodiac rafts. We were fortunate that one of our cruises had been chartered and with several descendants/relatives from the
-
.⏎Still an architectural marvel, the Seaside is today a complicated monument to both genuine attempts at medical innovation.⏎Know Before You Go⏎Vehicles should be left at the parking lot or grassy area before the gate. The only way to enter
-
th-18th, 2017, Vicenza, Italy)<|endoftext|>- Welcome to Botswana⏎- Where To Go⏎- What To Do⏎- Where To Stay⏎- Flora and Fauna⏎- Visitor Information
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
- L6:2660 excites
- L6:3062 suppresses
- L6:3001 suppresses
- L8:1720 suppresses
- L7:529 suppresses
- L6:2824 suppresses
- L6:2196 excites
- L6:2816 suppresses
- L7:1957 suppresses
- L7:102 suppresses
- L5:3030 weakly suppresses
- L4:2271 weakly excites
- L8:2850 weakly excites
- L3:659 weakly suppresses
- L8:775 weakly excites
- L6:1768 weakly suppresses
- L5:1155 weakly suppresses
- L7:676 weakly suppresses
- L7:1687 weakly excites
- L8:1456 weakly excites
- L7:581 weakly excites
- L4:2013 weakly suppresses
- L1:329 weakly excites
- L2:45 weakly excites
Feeds later layers
- L10:2368 suppresses
- L16:959 suppresses
- L22:2121 suppresses
- L22:1490 suppresses
- L23:2050 suppresses
- L22:2670 suppresses
- L10:1996 suppresses
- L21:3021 suppresses
- L19:1834 suppresses
- L23:1733 suppresses
- L20:2189 suppresses
- L17:2803 suppresses
- L10:2758 suppresses
- L11:3012 suppresses
- L10:2774 suppresses
- L21:1513 suppresses
- L21:3052 suppresses
- L22:665 suppresses
- L21:965 excites
- L26:2866 suppresses
- L19:2263 suppresses
- L10:2245 weakly suppresses
- L20:2226 weakly suppresses
- L22:1998 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
- 孀
- Bundesliga
- 酪
- .createClass
- 榄
- centage
- ificial
- parate
- inputData
- newData
Suppresses
- visit
- visits
- visited
- 游览
- visitor
- tours
- visiting
- visit
- 访
- tour