'that' after 'the way' or 'the first time'
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. 4 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 'that' when it follows phrases like 'the way' or 'the first time', as in 'the way that decimals are written'. Turning it up did not produce a clear pattern in the continuations.
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.
-
before the answer.⏎Subtracting money is easy, too, so long as it is once more remembered that money has to be written and subtracted the way that decimals are written and subtracted. Dollars go to the left of the decimal point
-
happy to report that the seeds that were transported to the moon have sprouted! While plants have grown on the International Space Station before, this is the first time that any biological material has grown on the moon. Why is this so important? Because
-
3 August 2012⏎Have you ever stepped out into a narrow city street and began to cough? You may have been a victim of the way that 'street canyons' trap polluted air, keeping harmful substances like nitrogen dioxide
-
study was posted online Dec. 29 in advance of print publication in the journal Proceedings of the National Academy of Sciences.⏎While this is not the first time that microbes in the human intestinal tract have been sequenced as a community, this is
-
should be able to find references or spell the word the way that it sounds.⏎Students should be writing for a variety of purposes.<|endoftext|>The Solar
-
harmful ones are bred out. But the “adaptive” nature of the changes natural selection (praise be upon it) wreaks may not be the only way that complexity grew.⏎Protein flaws called “dehydrons” may have made proteins
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: what is the value of the word, what is the value of the word, what is the value of the word, what is the value of the word, what is the value of the word, what is the value of the word,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
first author, Dr Damian Scarf of the University of Otago’s Department of Psychology, they performed this feat by tracking the statistical likelihood that “bigrams,” letter pairs such as “EN” and “AL,” were more likely associated with
-
- The messages contained within⏎- Illustrates human behaviour in any given situation and/or predicament⏎- Gives rules on life and living⏎- The romance of the narrative⏎- You have heroes
-
itself because its already in your mind, because if it does completely undo itself then there’s no memory of that experience anymore. People need to keep these memories because; these memories or remembered experiences make you wise and give you knowledge. Either congenial
-
]⏎- In case of other than assigned generator class, suppose if our generator class name is Increment means hibernate it self will assign the primary key id value into the database right [ other than assigned generator, hibernate only used to take care the
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
- L3:115 suppresses
- L3:2369 weakly suppresses
- L2:111 weakly suppresses
- L3:50 weakly excites
- L1:2493 weakly excites
- L3:186 weakly suppresses
- L1:1464 weakly excites
- L1:1064 weakly suppresses
- L3:1359 weakly suppresses
- L2:2349 weakly suppresses word pieces inside cinnamon, China, cabar
- L3:421 weakly suppresses
- L3:2858 weakly suppresses
- L2:704 weakly excites
- L0:1925 weakly suppresses
- L0:2903 weakly suppresses
- L3:691 weakly excites
- L1:2503 weakly suppresses
- L0:404 weakly excites
- L0:2414 weakly excites
- L2:2132 weakly excites
- L1:2990 weakly excites
- L0:1271 weakly excites
- L3:187 weakly excites
- L1:2358 weakly suppresses
Feeds later layers
- L7:114 excites
- L7:681 excites 'from' after prevent or stop
- L13:1439 excites end of document token
- L6:689 excites
- L5:1909 excites
- L8:392 excites
- L15:638 weakly excites "the" inside institution names
- L11:1272 weakly suppresses unclear: sentence ends in factual prose
- L14:1200 weakly excites
- L8:81 weakly excites
- L17:181 weakly excites
- L16:269 weakly suppresses
- L9:2354 weakly excites
- L27:2666 weakly suppresses
- L15:1554 weakly excites
- L17:359 weakly excites
- L13:929 weakly suppresses
- L17:1894 weakly suppresses
- L6:1524 weakly excites
- L6:244 weakly suppresses
- L18:2210 weakly suppresses
- L16:1258 weakly excites
- L25:1135 weakly excites
- L19:2058 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
- alties
- coming
- 时候
- Β
- ControlEvents
- getParameter
- aler
- 回事
- 些什么
- væ
Suppresses
- uo
- именно
- Product
- ioso
- Reflection
- }}
- Reflect
- 县政府
- Pope
- »