travel and journey verbs
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 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 responds to words about getting somewhere, such as travel, trip, return, home, back, and cross, especially where the text describes how long a journey takes. Turning it up gave text about distances and trips, though the match was loose.
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.
-
Illinois, will shoot particles six miles below the earth's surface to a new detector in Ash River, Minnesota, over 500 miles away. The whole trip takes about three milliseconds.⏎"It is a giant piece of plastic."⏎The Fermil
-
system is Proxima Centauri. The light of that star takes 4.25 years to reach Earth. That’s how long it takes light to travel 40,208,000,000,
-
of your ride whilst you are feeling fresh and your legs are strong. That way, assuming the wind doesn’t change, you can enjoy the tailwind on the return. This being said though, always use your judgement before heading out in the first
-
’t, and it took over 10 minutes to download the 48 second video.) Students can watch the programs on laptops, tablets
-
the potential to evade local weather problems that often impose untimely delays on traditional vertical rocket launches.”⏎Like The Hobbit, Stratolaunch will go there and back again. Image: Stratolaunch Systems Corp⏎By visiting different airports, it
-
operations. They must reduce the impact of the environment.<|endoftext|>5 COMMON CAUSES OF BRAIN FOG⏎You need to stop by the grocery on the way home from work for two gotta-have items.
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 list of things to do. The time of the day, and the temperature. It's time to go! The weather and the time of the week, and the time of the year, and the thing about the story... It's
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Whether you’re preparing a mathematics worksheet for class, or writing your scientific manuscripts, or typing up complex formulae in your research papers, JotterPad is here for you.⏎Insert TeX Equations With Ease⏎If learning
-
, raise public<|endoftext|>Homeowners across a large swath of Pennsylvania are reporting sightings of spotted lanternflies hatching from egg masses.⏎“Egg-hatching season is here, and that has some people on edge,” said Emelie Swackham
-
Google search Christmas traditions around the world, and ask your family to vote on a new tradition to adopt.⏎The prophet Isaiah looked forward to the arrival of the Prince of Peace, whose reign would lead to eternal shalom. Isaiah
-
of the city acting obscene or silly. Theatrical performances were also held at this festive time.⏎The Dionysia was meant as a celebration of spring’s return and the earth’s regeneration. As such, the festival included rituals dedicated to 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
- L6:1622 weakly excites
- L3:147 weakly excites
- L2:2371 weakly suppresses
- L4:2404 weakly excites
- L1:2966 weakly suppresses
- L6:2677 weakly excites
- L6:16 weakly excites
- L0:693 weakly excites
- L6:2921 weakly suppresses
- L5:1325 weakly suppresses
- L6:2549 weakly excites
- L4:3069 weakly suppresses
- L5:2845 weakly suppresses
- L5:2709 weakly excites
- L5:2058 weakly suppresses
- L0:3021 weakly suppresses
- L1:1057 weakly excites
- L5:2450 weakly suppresses
- L2:1190 weakly excites
- L5:812 weakly suppresses
- L5:2181 weakly excites
- L5:1276 weakly excites
- L3:3063 weakly suppresses
- L0:1002 weakly suppresses "the" right after an action verb
Feeds later layers
- L8:2030 suppresses
- L8:2957 weakly suppresses
- L8:1749 weakly excites
- L14:2429 weakly excites
- L9:2932 weakly suppresses
- L15:2691 weakly suppresses
- L18:2614 weakly excites
- L8:2716 weakly suppresses
- L9:652 weakly suppresses
- L16:2965 weakly suppresses
- L17:2571 weakly excites
- L11:1495 weakly excites
- L16:1313 weakly suppresses
- L8:2394 weakly suppresses
- L9:1748 weakly excites
- L17:2356 weakly suppresses
- L10:2261 weakly excites
- L8:2670 weakly suppresses comma after 'Unlike X' contrasts
- L14:2780 weakly suppresses
- L18:2277 weakly suppresses
- L9:2529 weakly suppresses
- L16:2043 weakly suppresses
- L10:3012 weakly excites
- L21:1826 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
- LEY
- agram
- >):
- '):
- -tracking
- lek
- 一段
- -side
- agrams
- -object
Suppresses
- обав
- 발표
- егод
- .notifyDataSetChanged
- eno
- Presence
- onas
- uppen
- 起来了
- .toCharArray