the 'de' syllable starting words
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. 26 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 the piece 'de' at the start of a word fragment, as in dehydrated, DeGrasse, devious, decongestants and deities. Steering up made the model drift into text about small things and feelings, and pushing down made it repeat the word brand.
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.
-
India is indeed hot and cruel, the long hot and often humid summer days, and the even more uncomfortable sultry, hot evenings in the capital leave us all dehydrated and exhausted. The heat, and the air conditioning, all take a
-
chances in<|endoftext|>About This Episode⏎What were early humans like? What did they eat? What did they do? On this episode of StarTalk Radio, Neil deGrasse Tyson and comic co-host Matt Kirshen investigate the diet of early
-
expression “talking about the birds and the bees” represents the most common version of our attempts to do this with some degree of biological accuracy. The jokes and devious explanations used by some parents also acknowledge our discomfort in dealing with the subject.⏎
-
body from stress – Our body needs to be de-stressed once in a while. Long work hours, unhealthy diet, and irregular
-
each step should be at least 10 inches, but no more than 12⏎Frequently Asked Questions⏎How Do You Build Stairs Between Two Decks?⏎There are a few ways to build stairs between decks, but the most
-
time to fix it, a lot of time,” he told MEE.⏎Among the shattered Assyrian treasures are two huge Lamassu (winged bull protective deities) from the ancient
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: brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand brand
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Snow White,” the beloved, first, full-length animated feature film, created by the Walt Disney Studios, is celebrating its 75th anniversary this year with a major exhibit at the Walt Disney Family Museum in San Francisco.⏎Called “Snow White
-
was occupied during the time of Moses and the Biblical Exodus.2 What do we make of this?⏎In “Kadesh-Barnea—In the Bible and on the Ground” in the September/October
-
algae greenhouses and the processing of algae – a plant with future.<|endoftext|>“I was forty years old when Moses the servant of the Lord sent me from Kadesh-Barnea to spy out the land; and I brought him an honest report.”⏎Where
-
desires, the right kind method to pursue happiness, and the correct plan for an excellent lifestyles.⏎Read or Download Aristotle for Everybody: Difficult Thought Made Easy PDF⏎Best greek &
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
- L24:637 excites
- L21:2130 suppresses
- L20:2890 suppresses
- L25:665 weakly suppresses
- L23:792 weakly excites
- L21:2084 weakly suppresses
- L17:1892 weakly excites
- L19:2971 weakly excites
- L19:1272 weakly suppresses
- L21:2743 weakly suppresses
- L20:180 weakly suppresses
- L10:1610 weakly excites
- L24:2875 weakly excites
- L24:2370 weakly excites
- L0:674 weakly suppresses
- L5:457 weakly suppresses 'a' after 'tell', 'bring' or 'much of'
- L16:126 weakly excites
- L4:137 weakly excites
- L20:901 weakly suppresses
- L21:1790 weakly excites
- L21:7 weakly excites mixed, weak code and reference hits
- L18:2744 weakly excites
- L16:139 weakly suppresses
- L1:1833 weakly excites
Feeds later layers
- L27:1658 weakly excites
- L27:1027 weakly suppresses
- L27:2611 weakly excites
- L27:2361 weakly suppresses
- L27:880 weakly excites
- L27:1628 weakly suppresses
- L27:2597 weakly suppresses
- L27:744 weakly excites
- L27:2201 weakly suppresses
- L27:2838 weakly suppresses
- L27:625 weakly excites
- L27:2848 weakly excites
- L27:612 weakly excites
- L27:738 weakly excites
- L27:925 weakly suppresses
- L27:2080 weakly excites first piece of a split name
- L27:1158 weakly excites
- L27:1587 weakly suppresses
- L27:681 weakly excites
- L27:850 weakly excites
- L27:185 weakly suppresses
- L27:1567 weakly suppresses
- L27:2072 weakly suppresses
- L27:523 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
- [++
- .annotations
- 篾
- glue
- 注
- edis
- edes
- s
- 副局长
- 独角
Suppresses
- brand
- Brand
- Brand
- BRAND
- .brand
- -brand
- brand
- 品牌的
- _brand
- brands