"well" before known or documented
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. 18 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 the word "well" when it comes before words like "known", "documented" or "studied", as in "well-known" or "well documented". It hardly reacts to anything else. Turning it up or down gave unrelated text, so the steering does not show what it does.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
X-Rays below.⏎You can read a full overview at Wikipedia here.<|endoftext|>Aloe vera has long been considered as one of Mother Nature’s miracle plants, well-regarded for its healing, antioxidant, anti-inflammatory, and moisturizing properties,
-
00.⏎In an attempt at transparency and regulation, Hungary began licensing sex workers in September 2007. Critics argued that such licensing<|endoftext|>It’s well understood that the digital divide disproportionately affects people living in regional Australian communities. Remote Aboriginal
-
involving 22 pairs of identical twins, it was discovered that following a vegan diet can improve cardiovascular health within a span of just eight weeks.⏎Although it’s well-known that eating less meat improves cardiovascular health, diet studies are often hampered by
-
. When every other student bombards him with questions and raises hand in every micro second, the teacher sits down with head in his hands.⏎· It is again well known that how extensively this method uses the support of online learning. This may involve
-
skype, artificial intelligence, and let’s not forget insect cyborgs. Yes, that is correct, scientists are well at work at creating an insect that is part machine and can carry out the demands
-
gatherings or stressful work projects to stifle the symptoms of anxiety. Anxiety can greatly affect your everyday activities. Symptoms of anxiety can start as early as childhood and last well into adult years.⏎There are multiple different types of anxiety disorders. These include generalized
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: how much of a difference this kind of “cognitive” process from a higher level of less than 1000 results in a true or “cognitive” process from a higher level of 1000? Also,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
BOUT THE SERIES: The Very Short Introductions series from Oxford University Press contains hundreds of titles in almost every subject area. These pocket-sized books are the perfect way to get ahead in a new subject quickly. Our expert authors combine facts,
-
the help of these easytouse reading comprehension cards! Your students will build critical reading skills by selecting the meanings of metaphors, similes and more.⏎- Perfect for center activities, homework and filling downtime between lessons⏎- Build critical reading skills
-
and recruiters who can help them reach their goals faster than they ever thought possible. And, of course, Guru99 notes that online job boards function as the perfect resource for finding work as a freelancer.⏎Finally, thanks again largely to technological advancement
-
efficiencies. It’s the greatest feeling to be pushing the envelope in electrical cars. Plus, everyone on the team has a passion for speed, so it’s the perfect project for us.”⏎Just as important as speed is the fact that electricity is renewable
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
- L4:1313 excites
- L16:1732 suppresses
- L0:1707 excites
- L3:2411 suppresses
- L5:244 suppresses
- L3:2823 suppresses
- L16:282 excites
- L2:1724 weakly suppresses
- L5:197 weakly excites
- L8:2899 weakly suppresses
- L16:374 weakly excites
- L16:64 weakly suppresses
- L16:1066 weakly suppresses
- L6:7 weakly suppresses
- L17:650 weakly excites
- L3:2037 weakly excites
- L5:623 weakly excites
- L1:262 weakly suppresses
- L3:1857 weakly excites
- L8:2477 weakly suppresses
- L15:89 weakly excites
- L17:337 weakly suppresses
- L2:75 weakly suppresses
- L0:1879 weakly excites
Feeds later layers
- L21:798 excites
- L25:1111 excites
- L24:1771 weakly suppresses
- L21:1999 weakly excites 'well' in compound adjectives
- L20:462 weakly suppresses
- L25:2879 weakly excites
- L19:994 weakly excites
- L19:2330 weakly excites
- L23:1459 weakly excites
- L19:2201 weakly suppresses
- L20:1064 weakly excites
- L24:2826 weakly excites
- L22:557 weakly suppresses
- L21:975 weakly excites
- L24:2131 weakly suppresses
- L27:1459 weakly excites
- L24:2914 weakly suppresses
- L23:1592 weakly excites
- L23:1194 weakly excites
- L24:1067 weakly excites
- L27:1318 weakly suppresses
- L20:2666 weakly excites
- L26:2748 weakly suppresses
- L26:1492 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
- -case
- luck
- behaved
- -quality
- -sh
- Luck
- quality
- fortune
- aub
- -su
Suppresses
- cznie
- electorate
- difference
- uncompressed
- StackSize
- versa
- irement
- っくり
- plier
- onComplete