'as' in 'just as'
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. 1 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 word 'as' when it comes straight after 'Just', as in 'Just as X, so Y'. The signal is fairly weak compared with the strongest neurons. Turning it up did not visibly change the model's writing in a related way.
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.
-
to maintain vitality throughout life – from childhood through adulthood. Just as getting older brings changes in career, family and activity goals, it also brings physiological
-
obvious way to protect your eyes from the blazing Delhi sun. Just as you use a sunscreen with adequate SPF, you must use sunglasses which offer appropriate UV
-
in the Western world, was also beset by civil war at the time. Just as Cleopatra was preparing to attack her brother with a large Arab army, the
-
there a certain kind of green zealot insists that methane may be better than coal but rather than making a prudent transition we should leap into the freezing dark. Just as so many of them are irrationally hostile to the extremely low GHG,
-
old framework which the reactionary force always tries to⏎Just as revolution is inevitable in politico-economic relations, revolution is inevitable in culture. A
-
value — the analogies are in many ways quite valid, because there are definite parallels between computational and biological viruses, and the means of battling against them. Just as we humans can minimize the risks of ill health through prevention and curing, the same
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 been thinking about the impact of the world. I have been thinking about the impact of the world. I have been thinking about the impact of the world. I have been thinking about the impact of the world. I have been thinking
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
overdoses, they tried to use newer forms of opioids to treat the addiction to its
-
ens but this study shows that it may be more complicated than that - asthma may involve colonization of the airways by multiple bacteria," said study co-author Homer Boushey, MD, a UCSF professor of medicine in the division of Pulmonary
-
why is it that when confronted with compelling evidence that college pays a big financial dividend, so many young people still don’t get a college degree?⏎Heather Boushey and I argue that the short answer is that for a surprising share of college
-
the policy of coexistence has been implemented in the past ten years.<|endoftext|>BBB - A+ Rating - the best there is⏎An excerpt from www.HouseOfNames.com archives copyright © 2000 - 201
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
- L0:1301 strongly suppresses
- L0:178 strongly suppresses
- L0:1224 suppresses
- L0:498 suppresses
- L0:397 weakly suppresses
- L0:41 weakly excites
- L0:760 weakly suppresses
- L0:1 weakly suppresses
- L0:2252 weakly suppresses
- L0:1036 weakly excites
- L0:1010 weakly suppresses
- L0:2518 weakly suppresses
- L0:962 weakly suppresses
- L0:1388 weakly suppresses
- L0:497 weakly suppresses
- L0:2759 weakly suppresses
- L0:1947 weakly suppresses
- L0:93 weakly suppresses
- L0:953 weakly suppresses
- L0:799 weakly suppresses
- L0:2944 weakly suppresses
- L0:2497 weakly suppresses
- L0:162 weakly suppresses
- L0:1063 weakly suppresses
Feeds later layers
- L12:129 weakly excites unclear: mixed, mostly everyday health and advice prose
- L2:400 weakly excites
- L11:622 weakly excites
- L18:2593 weakly suppresses
- L8:803 weakly excites
- L11:1670 weakly excites
- L12:1825 weakly excites
- L2:2831 weakly suppresses
- L11:2812 weakly excites
- L13:1847 weakly excites
- L20:376 weakly excites
- L14:1945 weakly suppresses
- L6:2215 weakly suppresses
- L11:1786 weakly excites
- L15:494 weakly excites
- L21:2623 weakly suppresses
- L12:567 weakly suppresses
- L6:2820 weakly suppresses
- L24:2112 weakly suppresses
- L26:370 weakly excites
- L9:462 weakly excites
- L2:2617 weakly excites
- L6:2700 weakly suppresses
- L3:118 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
- amental
- =val
- antes
- naments
- etting
- mel
- ackers
- synthesize
- cis
- imitives
Suppresses
- eki
- under
- seeds
- hti
- breadcrumbs
- litter
- TL
- EPA
- 阱
- Majesty