the word 'tropical'
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. 5 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 lights up on the word 'tropical', as in 'tropical forests', 'tropical storms' or 'tropical deforestation'. Its negative side is weak and mostly on fragments of other words, so it is mainly a word detector. Turning it up or down did not give a clear, related change in the model's writing.
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.
-
unknown and unnamed by scientists as of yet. Insects, especially beetles, make up a large proportion of this undiscovered life on Earth.Experts in remote tropical countries' fauna such as the wilderness of New Guinea, Alexander Riedel of
-
1, 2, 3, and this remains the cornerstone of most biodiversity conservation strategies 4, 5, 6. “The effects of tropical deforestation on climate go well beyond carbon,” says Professor Deborah Lawrence, “[it
-
ination of Humanitarian Affairs (OCHA), which also warned that the death toll could rise in the coming days. The cyclone has been downgraded to a tropical storm and is expected to affect other southern African countries, including southern Zimbabwe, northern
-
. 1. The Issue The …⏎BONN, Germany (June 05, 2014) – Programs and policies to reduce tropical deforestation, and the global warming emissions resulting from deforestation, are … May
-
will replenish the Federal Emergency Management Agency’s depleted funds in order to aid several states still reeling in the wake of widespread flooding, wildfires, tornadoes and tropical storms. But cleanup and recovery is only one part of the disaster cycle. Fortunately
-
ANE?⏎Hurricanes are tropical cyclones that occur in the Atlantic Ocean and have winds with a sustained speed of
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 a passion for the old world, and I am still waiting for it. The 20th Department of the People's Republic of the People's Daily. In the form of a newspaper, written in the style of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
commonly known as the draft. For those who are not aware, the draft is a system of conscription that enlists citizens into the military; the system is run using an online sign
-
ated five years later after the Soviet invasion of Afghanistan. Decades later the government website for the program lists the benefits of the new and “improved” conscription system. Stated, the draft is “the most equitable draft in history,”
-
itation: Formal references to statutes that describe where they are published. For instance, the citation 23 Vt. Stat. Ann. § 1185 tells us that this cited statute is
-
.⏎The project is eligible for School of Mathematics Doctoral Training Grant funding - please contact us for more information.<|endoftext|>História (S
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
- L1:1372 excites
- L3:1436 weakly suppresses
- L3:117 weakly excites
- L4:97 weakly suppresses
- L4:1795 weakly suppresses "at least" phrase
- L2:1799 weakly excites
- L4:25 weakly excites
- L4:42 weakly suppresses
- L4:1640 weakly excites
- L3:142 weakly suppresses
- L2:2903 weakly suppresses
- L0:1650 weakly excites
- L3:1981 weakly excites
- L4:23 weakly excites
- L3:672 weakly excites
- L1:642 weakly suppresses
- L2:2915 weakly suppresses
- L0:1960 weakly suppresses unclear: mixed medical and harsh words
- L4:1246 weakly suppresses
- L2:319 weakly excites 'of' after 'pattern' or 'patterns'
- L2:1756 weakly excites
- L0:1909 weakly excites
- L0:2799 weakly suppresses
- L2:2131 weakly excites
Feeds later layers
- L10:2035 suppresses
- L10:2514 excites
- L10:1705 excites
- L9:116 excites
- L9:209 excites
- L8:2052 excites
- L7:2370 weakly excites
- L6:89 weakly excites
- L8:2135 weakly suppresses
- L16:1147 weakly excites
- L19:1400 weakly suppresses
- L10:2533 weakly excites
- L19:783 weakly excites
- L21:599 weakly suppresses
- L10:1316 weakly excites
- L9:845 weakly excites
- L20:2861 weakly excites
- L20:1196 weakly suppresses
- L6:2164 weakly excites second half of multi-word proper names
- L6:1198 weakly excites
- L15:1553 weakly excites
- L17:1042 weakly excites
- L12:773 weakly suppresses
- L26:1950 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
- юр
- Crime
- greedy
- observational
- 很长
- avery
- Crime
- 庋
- Observation
- 忒
Suppresses
- atory
- ication
- ipated
- utory
- ipation
- illation
- cene
- 冻结
- bole
- aphore