'the' before a noun mid-sentence
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. 16 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 responds to the word 'the' in the middle of a sentence, especially after a preposition or verb, where a noun phrase is about to start, as in 'close to the skin' or 'away from the previous point'. It turns negative on 'the' in phrases like 'players of the official game'. Turning it up produced repetitive filler and loops rather than clear meaning.
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.
-
:⏎Teaching Strategies and Tips⏎Use this lesson to show how exponential functions and geometric sequences are related.⏎- In a geometric sequence, terms are found by multiplying the same constant to the previous term.⏎- For exponential functions, when the input variable
-
this by multiplying the number by the maximum possible mark. Once you’ve determined the maximum, you can also divide it by the total marks. This will give you the percentage of a given number. In addition to finding the actual value of a number
-
function that is part of the standard library (and uses the length of the data as the seed). It is quite insecure in a cryptographic sense, but since an educated cr
-
Web page: http://www.wider.unu.edu/⏎More information through EDIRC⏎Please report citation or reference errors to , or , if you are the registered author of the cited work, log in to your RePEc Author Service
-
visit our Prevention page. Briefly though:⏎- Use fine-point tweezers or a special tick-removing tool. Grasp the tick as close to the skin as possible. If you don’t have tweezers, protect your fingers
-
2.5 and the learning rate is 0.01, then the gradient descent algorithm will pick the next point 0.025 away from the previous point.⏎If you pick a learning rate that is too small, learning will
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: I need to be more of a "dualist" in the way of the old days of the "soul of the people" of the firsts. (if we are not going to get the "firsts" of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Spear & Sons Limited. Our scrabble word finder and scrabble cheat word builder is not associated with the Scrabble brand - we merely provide help for players of the official Scrabble game. All intellectual property rights to the game are owned by respective
-
it, “Cabby.” This is Cabby’s story:⏎To stay tuned on what Fant, and the rest of the Garden’s conservation scientists are doing, check out the latest news at chicagobot
-
iconic. Without it, Simon &- Garfunkel would not feel nearly as groovy. And just try⏎picturing Woody Allen’s Manhattan without that seeing him and Diane Keaton silhouetted against its criss-crossing
-
persons of another such description are not made subject or are accorded privileges or advantages<|endoftext|>Tom Standage, digital editor at the Economist, author of A History of the World in 6 Glasses and Writing on the Wall: Social Media - The First
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
- L6:61 suppresses
- L8:636 suppresses
- L9:24 excites "the" before a named organisation or title
- L5:125 excites "of" after all three, both, ten
- L12:435 suppresses
- L4:52 excites
- L15:470 suppresses the word 'this' after a preposition
- L7:2164 suppresses period inside country abbreviations
- L3:344 excites
- L7:111 suppresses
- L9:82 suppresses
- L11:1168 excites
- L10:167 suppresses
- L4:968 excites
- L10:29 suppresses
- L14:169 suppresses
- L14:2291 suppresses
- L0:26 suppresses
- L0:774 suppresses
- L5:1747 suppresses
- L14:2846 suppresses
- L13:975 excites
- L5:1172 excites
- L3:291 excites
Feeds later layers
- L17:1154 weakly suppresses
- L23:750 weakly excites
- L25:542 weakly suppresses
- L27:2391 weakly suppresses
- L27:829 weakly excites
- L18:2670 weakly excites
- L18:1976 weakly suppresses
- L18:1354 weakly excites
- L18:1399 weakly suppresses
- L21:2665 weakly excites
- L19:2163 weakly suppresses
- L17:1638 weakly suppresses
- L19:518 weakly suppresses
- L17:1336 weakly excites
- L18:2926 weakly excites
- L21:713 weakly excites
- L26:927 weakly suppresses
- L22:2451 weakly excites
- L19:1628 weakly excites
- L26:706 weakly excites
- L18:517 weakly suppresses
- L22:1659 weakly excites
- L21:847 weakly excites
- L20:2161 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
- 崮
- Users
- spons
- forEach
- herits
- 都市报
- poi
- 蚓
- AMPL
- sponsored
Suppresses
- ,{⏎
- например
- rí
- -basic
- 门
- <![
- <![
- <(
- //{⏎
- andbox