"your" addressing the reader
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. 8 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 "your" when a sentence speaks directly to the reader, as in "If your kitten..." or "Is your person...". It reacts in the opposite direction to third-person possessives like "its", "his" and "'s". Turning it up did not produce obviously related text.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
Kitten? How Do I Tell a Kitten’s Age?⏎If your kitten is yet to get its adult teeth, it is relatively easier to guess its
-
cancer. Whether triggered by an event or by one's own biochemistry, depression involves changes in the brain's neurotransmitter system. If you have ever had your mood altered by a drug, you know that how you feel is not simply a
-
in the Bible⏎In the morning they arose early and exchanged oaths; then Isaac sent them away and they departed from him in peace.⏎And if one man's ox does damage to another man's ox, causing its death, then the living
-
by By Dr. Carolyn K. Stratz⏎Q: My kids get their vaccinations - what about me?⏎A: Back-to-school time doesn't just mean
-
to the death over territory or a superior shell. Impact injuries, such as falls and being dropped, can lead to death.⏎What happens when crab die?⏎If your hermit crab dies, it will begin to decompose – and its
-
sports require an exorbitant amount<|endoftext|>Guess Who? Identity theft?⏎“Does your person have blue eyes?” “Does your person have massive eyebrows?” “Is your person wearing a hat?” “Is your person Frank?”⏎I was playing Guess
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: I have an old idea of building a model to simulate the environment of the future. To achieve this, I have an old idea of building a model to simulate the environment of the future. To achieve this, I have an old idea of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, „Washington Allston brought his enthusiasm for art and for romanticism into the very environment that needed them most and cared for them least. That was Allston's glory. That was Allston's tragedy.⏎Allston set up a studio in
-
– a majority of the garbage is comprised of tiny bits of plastic, barely visible to the human eye. Due to the small size of much of the garbage and its location floating 10 feet beneath the ocean’s surface, the garbage patch does
-
standstill, leaving, for a short period, the land open for God’s people to take over.⏎Special thanks to Dr. Robert L Diss Jr. for his materials on this
-
astronauts Jim McDivitt and Ed White — conducted scientific experiments, including photography of Earth’s weather and terrain, for the remainder of their four-day mission following White’s historic first American
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
- L4:217 suppresses
- L6:1233 suppresses
- L7:111 suppresses
- L5:1747 suppresses
- L5:125 excites "of" after all three, both, ten
- L4:300 excites
- L5:2275 excites
- L6:972 excites "well" after "how"
- L7:93 excites
- L5:53 excites
- L4:2853 excites
- L3:1318 weakly suppresses
- L4:2473 weakly excites
- L7:8 weakly excites
- L6:1718 weakly suppresses
- L7:2164 weakly suppresses period inside country abbreviations
- L4:1471 weakly excites
- L1:2550 weakly suppresses
- L6:1195 weakly excites
- L7:304 weakly excites
- L4:698 weakly suppresses
- L7:1853 weakly suppresses
- L1:2074 weakly suppresses
- L5:533 weakly excites
Feeds later layers
- L14:313 excites
- L9:1903 suppresses
- L11:3015 excites
- L15:448 excites
- L9:143 excites
- L15:2005 excites
- L16:940 suppresses
- L10:815 suppresses
- L14:1601 weakly excites
- L14:2150 weakly excites
- L19:238 weakly suppresses
- L16:1382 weakly excites comma after an introductory phrase
- L12:238 weakly suppresses
- L9:266 weakly excites
- L13:1099 weakly suppresses
- L13:1376 weakly excites
- L17:2792 weakly suppresses
- L15:1234 weakly suppresses
- L9:288 weakly excites
- L17:1194 weakly suppresses
- L22:552 weakly suppresses
- L17:1959 weakly suppresses
- L16:559 weakly suppresses weak, mixed: verbs and adverbs in claims
- L13:909 weakly suppresses 'to' before a verb after a noun
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
- liğinde
- enough
- 字号
- 好几个
- tle
- keyed
- Ngoại
- TimeString
- somebody
- cards
Suppresses
- eliness
- ament
- uel
- ugs
- \",⏎
- amel
- agascar
- �
- lust
- âm