the word COVID before -19
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 lights up on the word COVID when it comes right before '-19', as in news about the pandemic. Turning it up made the writing wander into odd capitalised words and repeated phrases.
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.
-
their essay and learned to focus on points in their lives, rather than the entirety of it.<|endoftext|>COVID-19 Vaccine Enters Final Test Phase⏎The COVID-19 pandemic seemingly has no expiration date. At its onset, we thought
-
public – as social distancing restrictions are eased – can help us further reduce the spread of COVID-19 that we’ve achieved by staying at home,” said Bill Hayes,
-
pandemics such as COVID-19. Suppression can work if the outbreak is small and you’re
-
ADA), to protect workers who contract COVID-19. These guidelines were set because the impact of COVID-19 on the mind and body can meet the ADA's definition of a
-
countries and we expect it to soon be the dominant COVID-19 strain circulating worldwide.⏎“The world is watching in real-time as the COVID-19 virus continues to change and become more transmissible.”⏎Kindly follow
-
that people should be well informed about the disease-causing virus and the ways through which it spreads the disease. The only easy and effective way to protect yourself from COVID-19 is by using an alcohol-based massage or washing your hands frequently,
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 5 of 5, 5, 5, 5, 5, 5. and 5, 5, , and, 5. 5, and, 5,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
audio⏎OAK RIDGE, Tenn., Dec. 7, 2005 —
-
a maximum of 100 mg/day. The American Congress of Obstetricians and Gynecologists recommends that first-line treatment of nausea and<|endoftext|>PITTSBURGH — Do you have a daily indulgence in your diet? Even in
-
was reduced to a province within the Middle Assyrian Empire.⏎ Some historians place the Kassites and Hurrians within the bounds of Mitanni.⏎The KASSITES – c. 1184-763⏎When Matanni
-
images from the receiving ends of the bundled fibers to the surface (the eyeballs), creating the illusion of animated eyes. In addition to moving eyes, PAPILLON can project expressive symbols (such as dollar signs) in place of pupils to
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
- L4:533 strongly suppresses
- L5:35 strongly suppresses
- L7:128 strongly suppresses
- L3:627 strongly suppresses
- L6:2269 strongly suppresses
- L9:1737 suppresses
- L1:1488 suppresses
- L0:76 excites
- L1:1260 suppresses
- L1:1124 excites
- L0:704 excites
- L0:2979 suppresses
- L1:2221 excites
- L3:2329 excites
- L5:281 excites
- L3:1479 suppresses
- L6:430 suppresses
- L2:2051 excites
- L2:1830 suppresses
- L0:1904 suppresses
- L6:861 excites
- L7:524 suppresses
- L0:2229 suppresses
- L2:2174 excites
Feeds later layers
- L16:584 strongly excites
- L11:631 strongly excites
- L18:373 strongly excites
- L22:651 strongly excites
- L20:896 excites mid-word fragments in capitalised titles
- L22:1126 excites
- L15:257 excites
- L11:332 suppresses
- L11:322 excites
- L24:1991 excites
- L18:1733 excites
- L20:562 excites
- L18:599 excites
- L25:1750 excites
- L24:980 excites
- L20:159 suppresses
- L26:1531 excites
- L21:533 excites
- L27:782 excites
- L17:1747 suppresses split tech names and acronyms
- L27:1360 excites
- L22:212 excites
- L17:1622 excites
- L14:642 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
- _BEGIN
- MonoBehaviour
- solid
- linux
- hipster
- peninsula
- addtogroup
- uga
- ',⏎
- ushort
Suppresses
- 第八
- arith
- -o
- 第五
- ение
- ervative
- fäll
- 侬
- 制造
- encial