the digit 1 in COVID-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. 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 responds to the number 1 in the word COVID-19, which is the token that comes right after 'COVID-'. It barely reacts to anything else. Turning it up or down changed the model's writing in ways that have nothing to do with COVID, so the steering results don't reveal much.
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.
-
facilitating a learning environment, personal growth and identifying students needing attention. Let’s discuss why healthy learning should be brought together for the betterment of students.⏎COVID-19 Has Transformed Traditional Learning Methodologies⏎Among the challenges brought by the pandemic
-
will have come up with a workable topic for 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
-
from COVID-19 is by using an alcohol-based massage or washing your hands frequently, and not touching your mouth, nose, and eyes.⏎The COVID-19 virus spreads mainly through dewdrops of saliva or discharge from the nose when the
-
Hopkins researchers recently received a $195,000 Rapid Response Research grant from the National Science Foundation to, using machine learning, identify which COVID-19 patients are at risk of adverse cardiac events such as heart failure, sustained abnormal
-
COVID-19 diagnostic based on micro/nano technology and how to tackle the above-mentioned
-
COVID-19. This is an old drug for the new virus. At least since
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: the first is a simple approach of the first, which is the recursive approach. So, I think, the problem with the recursion, and the problem with the list, and the problem with the list, and the problem with the list,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
012, https://www.hsph.harvard.edu/nutritionsource/vitamins/.⏎- Betta Health, VIC Government, https://www.betterhealth.vic.gov.au/health/healthyliving/antioxidants⏎
-
Enchanted Learning website for:|⏎Copyright ©2002-2016
-
a highly promising new approach for the development of drugs to treat pneumococci.⏎As they report in the journal Angewandte Chemie, the scientists copied the choline architecture of the pneumococ
-
95% of all workloads will be completed in the cloud.⏎Cloud computing initiates a grand shift from the old traditional IT management systems to new and improved digitally-powered administration. Some benefits of cloud computing include:⏎- Reduction of capital costs
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
- L3:270 excites
- L4:2141 excites
- L4:2286 excites
- L4:1362 excites
- L4:2046 weakly suppresses
- L4:2342 weakly excites
- L4:637 weakly suppresses
- L4:2240 weakly suppresses
- L1:1433 weakly excites
- L1:2789 weakly excites
- L1:2622 weakly excites
- L4:2379 weakly suppresses
- L4:2541 weakly excites
- L1:2760 weakly excites
- L3:1561 weakly excites
- L3:1860 weakly excites
- L1:191 weakly suppresses
- L1:645 weakly excites
- L4:2759 weakly suppresses
- L3:438 weakly suppresses
- L0:256 weakly excites
- L0:2608 weakly excites
- L2:1567 weakly excites
- L3:583 weakly excites
Feeds later layers
- L7:2512 suppresses
- L11:2747 weakly suppresses
- L8:1559 weakly excites
- L11:691 weakly excites
- L14:915 weakly suppresses
- L10:2413 weakly suppresses
- L7:573 weakly suppresses
- L27:2678 weakly suppresses
- L22:1279 weakly suppresses
- L21:50 weakly excites comma after 'So' or intro phrase
- L15:1069 weakly suppresses
- L12:1497 weakly excites
- L8:774 weakly excites
- L20:2598 weakly suppresses
- L8:1475 weakly excites
- L8:443 weakly excites
- L25:1538 weakly suppresses
- L9:916 weakly suppresses fractions and ratios in maths
- L16:2812 weakly suppresses
- L26:814 weakly excites
- L12:1802 weakly excites
- L21:1400 weakly suppresses
- L12:1319 weakly suppresses
- L7:2467 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
- дор
- freezer
- marks
- rew
- /
- đạo
- asurable
- ành
- marks
- assage
Suppresses
- onica
- kaz
- oldValue
- 今天
- 今天我们
- tica
- сан
- ypes
- alta
- akens