the word 'following' after 'the'
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. 0 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 'following' in phrases like 'which of the following' or 'read the following', the kind of wording used in quizzes and instructions. When it was turned up, the model's text drifted into repeated phrases rather than anything about quizzes.
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.
-
options contract has the following properties or attributes:⏎- Expiration Date⏎- Symbol or Underlying⏎-
-
-choice question for which there is no single correct answer but for which there are perhaps more justifiable answers and less justifiable answers. Something like “Which of the following motivations best explains so-and-so’s actions in such-and-such novel?” Instead
-
<|endoftext|>Recall that the conjugate base of an acid is simply the compound formed when an acid loses a hydronium ion (H+).⏎Which the following represent conjugate acid–base pair? For the pair that is not conjugate
-
:37⏎Exercise7. Read the following sentences, underline the main verb and state the verb: Active or Passive.⏎1
-
⏎A. Modern Model⏎B. Classful Model⏎C. Classless Model⏎D. Heterogeneous Model⏎6-5. Which of the following is correct in CIDR?⏎A. Class A includes Class B network⏎B
-
. 4 bits⏎B. 8 bits⏎C. 16 bits⏎D. 32 bits⏎6-7. Which of the following is reliable communication?⏎D. All of them⏎6-8. What is
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 have been thinking about lately, and I have been thinking about lately. So I am thinking about lately. So I am thinking about lately. So I am thinking about lately. I have been thinking about lately. So I am thinking about
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
5 eroding from a bank at approximately⏎25 cm below the current terrain⏎surface (see in-situ photo below).⏎Its front surface,⏎
-
one ocean sitting below the familiar Ice I found on Earth and on top of different high-pressure ices.⏎
-
range from shallow (2” to 3” below the surface) to deep bands at 6” or lower. Deep applications tend to
-
facsimile reprint of a 1901 variation by means of the Macmill<|endoftext|>This question seems like such a common occurrence. It most always follows the realization that we have a visual impairment. Honestly, it IS a fair question.
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
None: this is the first layer.
Feeds later layers
- L12:1289 weakly excites
- L15:585 weakly suppresses
- L17:908 weakly excites
- L11:1267 weakly suppresses
- L11:2552 weakly excites
- L22:780 weakly suppresses
- L14:2266 weakly excites
- L20:2017 weakly excites
- L6:487 weakly excites
- L15:721 weakly excites
- L6:539 weakly excites
- L23:2293 weakly suppresses
- L15:2582 weakly excites
- L17:1965 weakly suppresses
- L16:1979 weakly excites
- L14:1701 weakly excites
- L8:2999 weakly suppresses
- L14:2539 weakly suppresses
- L20:1532 weakly excites
- L26:2333 weakly excites
- L18:2270 weakly suppresses
- L22:365 weakly suppresses
- L26:2478 weakly suppresses
- L25:1602 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
- yk
- yne
- imens
- 的操作
- ysize
- ayscale
- enso
- nop
- implement
- iphertext
Suppresses
- solitude
- 决
- libero
- treffen
- _None
- 慰
- 流出
- 流水
- الدنيا
- ance