slot before a year in dates
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. 11 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 tends to light up on the spot just before a year, such as the space between 'November 13,' and '1850'. It also fires on a few unrelated word positions. Turning it up gave rambling text that did not clearly relate to dates, so its effect on writing is hard to read.
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.
-
Countess of Rosebery, died, at Dalmeny House, in Scotland, at the age of 39. As the only child of Meyer
-
daily amounts of vitamin D are:⏎19 to 70: 600 IU<|endoftext|>Stevenson was born on Nov. 13, 1850.
-
the top banner and click on the three dots or parallel lines. Scroll down to the subscription form and enter your information.)<|endoftext|>Born on December 28, 1926 to Reverend Wade Hampton McKinney and Ruth Berry McKinney
-
Vogela (later called Fanny) and Loeb (later called Levi). Levi, the youngest of the family, was born on February 16, 1829 in the Bavarian village of Buttenheim.⏎Life in
-
butterfly and sweetheart, respectively, in Visayan) were born in our shipping pens area on December 6, 2012, and they have quickly become two of my absolute favorite charges
-
you add in the details. They help you form action steps and timelines to reach your goals and bring you the success you desire.⏎<|endoftext|>Born May 3, 1791, in St
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 person is the person who is in the room. The person who is in the room is the other person. The person who is in the room is the other person. The other person is the other person. The other person is the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
0.1 bel, so the decibel value LdB is LdB = 10 log10 (P1/P0).⏎Define the mean squared error (MSE) between two monochromatic images, where one image
-
is considered to be an approximation of the other. The MSE can be described as the mean of the square of the differences in the pixel values between the corresponding pixels of the two images.⏎Express
-
liturgy celebrated?⏎- Where is the liturgy celebrated?<|endoftext|>A particle is moving along the parabola 4y = (x + 2)^2 in such a way that its x-coordinate is increasing at a constant rate of
-
⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16 25.........⏎Basics: Divide integers to get integer outputs in all cases⏎D
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
- L10:937 excites
- L9:1940 excites
- L6:2423 suppresses
- L7:2233 weakly excites
- L0:857 weakly suppresses
- L9:2253 weakly suppresses
- L10:1673 weakly suppresses
- L0:2620 weakly excites
- L10:2399 weakly suppresses
- L7:1084 weakly excites
- L9:2295 weakly excites
- L2:1623 weakly excites
- L10:278 weakly suppresses
- L5:2672 weakly excites comma after 'As a [role]'
- L7:3036 weakly excites
- L7:1885 weakly excites
- L5:1763 weakly suppresses
- L9:2033 weakly excites
- L6:709 weakly excites
- L6:2259 weakly excites
- L4:1460 weakly suppresses
- L5:2719 weakly suppresses
- L7:2269 weakly excites
- L7:2137 weakly suppresses
Feeds later layers
- L12:2189 excites
- L13:1628 excites
- L12:3045 excites
- L17:1762 excites
- L13:1112 excites
- L12:887 weakly excites
- L13:2328 weakly suppresses
- L12:1854 weakly excites
- L15:1425 weakly excites
- L14:2935 weakly excites
- L12:576 weakly excites 'of' in fixed phrases like table of contents
- L13:1622 weakly excites
- L13:1001 weakly suppresses
- L14:1353 weakly excites
- L14:1265 weakly excites
- L27:2842 weakly excites
- L13:1513 weakly excites
- L16:1344 weakly excites
- L14:2169 weakly excites
- L17:2768 weakly excites
- L17:718 weakly excites
- L13:2963 weakly excites
- L22:956 weakly suppresses
- L15:2628 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
- аз
- adratic
- okedex
- thern
- ²
- 详
- ikk
- Quadr
- очно
- tık
Suppresses
- .Consumer
- ccoli
- .getCount
- getService
- maid
- 数目
- olding
- ировка
- 每一个
- _<