digit 1 in COVID-19 and 1941
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. 10 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 reacts to the digit 1 in COVID-19 and in years like 1941. Its opposite side lights up on words like climate and warming in discussions of climate change. Turning it up did not produce a visible theme in the output.
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.
-
the efficacy of its anti-malarial properties.⏎Currently, several trials which are being conducted around the world seem to indicate evidence to use this drug for COVID-19. I caution that these are preliminary results. In previous scientific publications it appears
-
a long time. And also, we are able to detect early disease,” Gouma said.⏎Researchers utilizing the breath test were able to identify COVID-19 in patients by observing how the virus reacts with oxygen, nitric oxide,
-
health of women throughout their lives. … A professional in midwifery is known as a midwife<|endoftext|>Clearly we are far from eradicating COVID 19 from our environment, but we do have the power to boost our own defenses
-
due to an underground coal-mine fire that’s been burning since 1962, which has made it too hazardous to live there; and Oradour-sur-Glane, France, a town that lost 600 lives and
-
dragged into war. In 1940 Japan with threats of force entered French Indochina. Finally, on December 7, 1941, Japan launched an armed attack on the United States, followed immediately by declarations of
-
of the cellular life forms that they infect; they turn the metabolism—the energy—of any cell that they infect into the reproduction of that virus. For COVID-19 we are fuel for the fire.⏎When
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: [Context] I'm writing a story about a character who is trying to do something specific things. I'm writing a story where a character is in a state of [X], and [X] is also a character with a specific name
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are increasingly being used for a range of biodiversity applications such as modelling the distribution of communities, ecological refuges, potential impacts under climate change, and biotic interactions (Araújo & Luoto,
-
it could become a test program for other coastal communities around the state.⏎Hawaii has rarely been directly hit by hurricanes. However, Lee writes in the bill that warming oceans increase the risk of hurricanes for the state. The bill estimates that if a
-
shift in the bacteria's range, he said. In addition, diseases such as cholera often spread following events like flooding, which may become more common with climate change, Morris said.⏎Other waterborne diseases can come from harmful algal blooms,
-
have been broken. The conditions have been so extreme that wildfires have broken out in Sweden and the UK.⏎Heatwaves are one of the most likely consequences of climate change. As the average global temperature rises due to higher levels of greenhouse gases, more
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
- L9:956 excites
- L9:271 suppresses
- L9:1906 excites
- L8:1545 excites
- L8:2230 suppresses
- L8:1469 suppresses
- L6:932 weakly excites
- L8:2342 weakly suppresses
- L8:1760 weakly excites
- L9:2155 weakly excites
- L8:2346 weakly suppresses
- L9:902 weakly excites covid and pandemic news words
- L2:2759 weakly suppresses
- L6:2050 weakly suppresses
- L5:951 weakly excites
- L5:2760 weakly excites
- L8:929 weakly excites
- L9:1299 weakly suppresses
- L6:1162 weakly excites
- L2:1732 weakly excites
- L1:1119 weakly suppresses
- L3:1212 weakly suppresses
- L2:2070 weakly suppresses
- L7:2619 weakly suppresses
Feeds later layers
- L13:2397 excites
- L11:1528 excites
- L16:602 suppresses
- L18:1961 suppresses
- L13:328 weakly suppresses
- L11:1287 weakly suppresses
- L19:1458 weakly suppresses
- L20:3060 weakly suppresses
- L14:1412 weakly suppresses
- L11:1103 weakly suppresses
- L16:1964 weakly excites
- L16:1551 weakly excites
- L11:1316 weakly excites
- L16:2254 weakly suppresses
- L14:147 weakly suppresses
- L11:1409 weakly suppresses
- L12:1227 weakly excites
- L17:1850 weakly excites
- L17:705 weakly suppresses
- L12:1226 weakly suppresses
- L11:1731 weakly excites
- L11:2032 weakly excites
- L11:1371 weakly suppresses
- L11:316 weakly suppresses
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
- 就说
- 堂
- amsung
- 琬
- arges
- zano
- argued
- atitude
- /Gate
- []=$
Suppresses
- 墟
- light
- -fix
- ib
- hash
- ios
- 净
- medium
- emat
- setting