'and' in 'more and more' pairs
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 fires on the word 'and' in phrases that repeat a comparison, such as 'more and more' or 'stronger and more united'. Turning it up did not produce that phrase and gave unrelated text.
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.
-
in 2010 on the promise of a “tory miracle”. He promised a “country that leaves the European Union” will emerge “a stronger and more united nation”. His party is led by Cameron and his government has pushed through
-
5-years-of-age, more and more opt not to drive at night. In addition to senior drivers, a significant
-
writing career, Vladimir Nabokov was frustrated by both the Freudian and behavorialist schools of psychology that reigned in his day, seeking a richer and more complex view of the mind than either provided. In a long, incis
-
1 Julkaistu artikkeli, soviteltu|<|endoftext|>Though it straddles the Arizona and Utah borders, Monument Valley is a symbol bigger and wider-ranging than either of the two states. Watch The Searchers and it’s
-
the largest in the kingdom. In the late fourteenth century<|endoftext|>With the ever growing popularity of Linux as a desktop operating system, more and more people are beginning to see it as a usable and valid alternative to proprietary software
-
“artificial intelligence.” And while it may sound like something out of a science fiction novel, the truth is that AI is very real and it is becoming more and more commonplace in our everyday lives.⏎Creating A Story With AI⏎So, what
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 answer to the question. If you can't get it, it will be included in the answer. If you can't get it, it will be excluded from the answer. (1) This is a true reflection of the actual problem
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a high likelihood of developing one or more of these common precancers if they spend time in the sun and live long enough. Location makes a difference: The closer to the equator you live, the more likely you are
-
cotangent as⏎Both of these values can be decimals. The range of cotangent also has no restrictions; the ratios are any and all numbers,⏎The closer you get to the values where⏎the smaller the number on the bottom of your fraction
-
ice. A new model, based on experiments in the laboratory that simulate salty seas, shows that the ocean and ice may be stacked up in multiple layers, more like a club sandwich.⏎Ice comes in different forms depending on pressures. "Ice I
-
the course of all food regardless of the time in which need to consume, so you can say desserts as such do not exist in Chinese food, the more close to these in Chinese food are the sweet dishes that are served.<|endoftext|>PREVIOUS :
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
- L4:656 excites
- L3:346 weakly excites
- L4:1584 weakly suppresses
- L0:1325 weakly suppresses
- L0:560 weakly suppresses
- L4:686 weakly suppresses
- L3:456 weakly suppresses
- L4:2751 weakly suppresses
- L2:2895 weakly excites
- L0:1924 weakly excites
- L1:1144 weakly excites
- L4:2264 weakly excites
- L1:2906 weakly excites
- L3:59 weakly excites
- L1:2970 weakly excites
- L0:529 weakly excites
- L2:220 weakly excites
- L0:165 weakly excites
- L1:1700 weakly excites
- L0:1671 weakly suppresses
- L2:3015 weakly suppresses
- L4:2117 weakly suppresses
- L3:177 weakly suppresses
- L0:3050 weakly suppresses
Feeds later layers
- L7:2157 strongly excites 'the' after 'like' or 'stay'
- L17:1962 suppresses
- L19:1716 suppresses
- L27:3027 excites
- L6:960 weakly excites
- L19:1065 weakly suppresses
- L15:1333 weakly excites
- L13:227 weakly excites
- L18:1175 weakly suppresses
- L16:3037 weakly suppresses
- L18:1309 weakly suppresses
- L8:2741 weakly suppresses
- L11:338 weakly excites
- L12:801 weakly excites
- L17:698 weakly suppresses
- L7:534 weakly suppresses 'much' or 'many' after 'not' or 'little'
- L18:3065 weakly suppresses
- L8:312 weakly suppresses
- L15:1301 weakly suppresses
- L10:1871 weakly excites
- L8:1074 weakly suppresses
- L11:1537 weakly suppresses
- L11:922 weakly suppresses
- L8:2694 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
- .Builder
- 之
- feit
- ania
- ActiveForm
- um
- ancies
- double
- .getDouble
- uman
Suppresses
- -than
- vib
- 倾向
- 亲
- ieg
- hog
- .Lib
- ARED
- elsinki
- Than