"quite" in hedged negations
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. 6 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 "quite" when it follows a negation or doubt, as in "not quite sure" or "can't quite work out." Turning it up made the model's text drift into unrelated topics, so the steering results don't show this pattern clearly.
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.
-
regiments to be raised from the population of California to help protect mail routes and the communications lines of the West. Neither Lincoln nor the U.S. War Department quite trusted the Mormons of the Utah Territory to remain loyal to the Union, in spite
-
I don't know how to review it. We're adding the slope at a single point to the slope over a big area, which is something I can't quite work out visually.<|endoftext|>The world produces more than enough food to feed everyone,
-
exit out of the loop, IE it will continue to call it self over and over until the program finally kills the heap/stack or something. I don't quite know what it is...⏎Also, even if it wasn't a static, it
-
u-substitution,⏎- we could say that du is equal to 5 cosine 5x.⏎- Now when you look over here, we don't have quite du there. We have just cosine of 5x dx--⏎- sorry,
-
ever wondered what a cobot is?⏎You might have come across this term “cobots” while searching for a solution to improve your production processes but weren’t quite sure what the term referred to. As you’ve probably noticed yourself, cobots
-
recognize the goodness of American foreign policy and the inherent democratic character of the United States. Bush increasingly appears somewhat like a frustrated Lyndon Baines Johnson who could not quite understand why “all” Americans did not recognize their prosperity during the democratic upheav
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 number of people who have been in the system. The number of people who have been in the system is equal to the number of people who have been in the system. Let me try to figure out the next possible thing to add.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
fought for⏎civil rights, child welfare and womens rights.<|endoftext|>How does Ray Bradbury portray technology in the story "There Will Come Soft Rains?"⏎2 Answers | Add Yours⏎The technology in the story is actually presented as the main character
-
Carl Ruetz⏎Page last modified July 14, 2009<|endoftext|>what does hamlet think about suicide? with textual support⏎2 Answers | Add Yours⏎Hamlet's thoughts on suicide occur at two places in the
-
in 6 bytes, with the last byte equal to 0.⏎October 3rd, 2010, 07:44 AM⏎We are not going to do you homework.⏎If you want our help, you
-
), and A[1:0] for ARM instruction fetches (nOPC=0 and MAS[1:0]=10).⏎Article last edited on: 2008-09-09 15:
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
- L3:255 excites
- L3:148 weakly suppresses
- L2:3004 weakly suppresses
- L5:2155 weakly excites
- L5:1803 weakly excites
- L0:2768 weakly suppresses
- L1:1299 weakly suppresses
- L3:52 weakly suppresses
- L5:1872 weakly suppresses
- L4:413 weakly excites
- L2:1988 weakly suppresses
- L2:2074 weakly excites
- L5:1380 weakly excites
- L5:866 weakly excites
- L5:1920 weakly suppresses
- L1:2669 weakly excites
- L0:2722 weakly excites
- L2:1081 weakly excites
- L4:2883 weakly excites
- L2:899 weakly suppresses
- L3:767 weakly suppresses
- L4:618 weakly suppresses
- L1:506 weakly excites
- L5:3027 weakly suppresses
Feeds later layers
- L7:1992 weakly excites
- L13:1958 weakly excites
- L23:1443 weakly excites
- L7:2773 weakly excites
- L7:1770 weakly excites
- L14:1083 weakly suppresses
- L14:812 weakly suppresses
- L9:2679 weakly suppresses
- L11:2552 weakly excites
- L9:1772 weakly suppresses
- L10:2908 weakly excites
- L17:1201 weakly suppresses
- L13:2488 weakly suppresses
- L16:2202 weakly excites
- L8:1854 weakly suppresses
- L17:2288 weakly excites
- L23:1452 weakly suppresses
- L8:2912 weakly excites
- L10:1758 weakly suppresses
- L14:431 weakly suppresses
- L8:2999 weakly suppresses
- L8:2355 weakly excites
- L15:505 weakly suppresses
- L9:2961 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
- amber
- ESH
- atches
- 뜬
- PropTypes
- exhaust
- *)&
- 昆
- computing
- domestically
Suppresses
- olution
- tür
- 褓
- alary
- lsx
- mdir
- 越高
- vio
- levation
- ians