unclear: asymptotes and electron loss or gain
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. 10 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 is mixed. It fires on the word 'asymptotes' in graphing lessons and on some web boilerplate, and it goes negative on 'lose' or 'gain' when electrons are described in chemistry. Turning it up or down did not give a clear theme in the generated text.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
(this will throw an Error in a future version of PHP) in /home/commons/public_html/wp-content/plugins/ultimate-google-analytics/ultimate_ga.php on line 524⏎GUJHS. 2007 March;
-
general shape of your graph so that you have a nice place to start.⏎To sketch the full parent graph of cotangent, follow these steps:⏎Find the vertical asymptotes so you can find the domain.⏎Because cotangent is the quotient of cosine
-
(in Central America)|⏎|Scientific Name||Lampropeltis triangulum gaigeae|⏎|Conservation Status||Not listed by IUCN|⏎When these snakes are young,
-
like with tangent. However, these asymptotes occur whenever the⏎The cotangent parent graph repeats every pi units. Its domain is based on its vertical asymptotes: The first one comes at 0 and then repeats every pi radians.
-
(carmelite) [x] st. terasea of the little flower [x] the st. francis [x] [Add Your Own Tags]⏎* Works in the collection are moved off view for many different reasons. Although
-
of the year, occurs every year between July 17 and August 24. The shower tends to peak around August 9-13. These meteor showers are really interesting to see, however if you live in an urban area being
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: This is a good article, this is a good article, and this is a good article. I have been thinking about, this is a good article, and this is a good article, and this is a good article. This is a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
configuration. The Alkali metals and Alkaline earth metals have one and two valence electrons (electrons in the outer shell) respectively. These elements lose electrons to form bonds easily, and are thus very reactive. These elements are the
-
almost never react, since they have eight valence electrons, which makes it very stable. The halogens, directly to the left of the noble gases, readily gain electrons and react with metals. The s and p blocks make up the main-group
-
1.176-1.709) and vomiting (p < 0.001, OR = 1.969, 95% CI = 1.618-2
-
10-10 m? Which of the following orbital designations are incorrect: 1s, 1p, 7d, 9s, 3f, 4f,
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
- L9:879 weakly excites
- L7:648 weakly excites
- L9:1015 weakly suppresses
- L5:2741 weakly excites
- L5:1015 weakly excites
- L2:35 weakly excites
- L0:16 weakly excites
- L9:1969 weakly excites
- L7:58 weakly suppresses
- L4:449 weakly suppresses
- L6:1011 weakly suppresses
- L1:2342 weakly excites
- L1:2864 weakly excites
- L8:1864 weakly excites
- L9:2678 weakly excites
- L5:1923 weakly excites
- L8:2299 weakly excites
- L4:1429 weakly suppresses
- L7:940 weakly suppresses
- L0:180 weakly excites
- L1:267 weakly suppresses
- L3:2974 weakly excites
- L2:876 weakly suppresses
- L4:1690 weakly excites
Feeds later layers
- L12:1854 suppresses
- L11:2019 weakly suppresses
- L24:1193 weakly suppresses
- L19:2997 weakly suppresses
- L12:2260 weakly excites
- L20:2808 weakly suppresses
- L11:703 weakly excites
- L23:2555 weakly suppresses
- L22:1282 weakly excites
- L13:1389 weakly suppresses
- L24:2547 weakly excites
- L14:2354 weakly suppresses
- L22:231 weakly suppresses
- L12:653 weakly suppresses
- L17:2618 weakly excites
- L23:431 weakly suppresses
- L12:1979 weakly suppresses
- L18:430 weakly suppresses
- L23:1241 weakly suppresses
- L24:1802 weakly excites
- L19:1960 weakly suppresses
- L25:2072 weakly excites
- L19:1757 weakly excites
- L11:2880 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
- .every
- Cad
- หลาย
- ably
- ';";⏎
- 一顿
- 루
- ఆ
- fully
- cps
Suppresses
- uhn
- icons
- ought
- zer
- idi
- rence
- ники
- avier
- ZW
- 松弛