'even' after 'but' or 'while'
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
The neuron fires on the word 'even' when it comes in the middle of a sentence after words like 'But', 'while', or 'such', as in 'But even when you want to write'. Turning it up didn't produce a clear theme, so the model's replies were mostly unrelated.
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.
-
seeking war with Iran – No!), this is crazy. First, he’s advocating for homosexually-active men to enter fake marriages. This is defending marriage? But even better, let’s take a quick look at Alexander’s marriage policies, from Ar
-
Gilly Hopkins, like most of Paterson's novels for young adults, addresses the difficulty of growing up in an unpredictable world where even friendship and family cannot be taken for granted. Eleven-year-old Gilly Hopkins has
-
11<|endoftext|>For many people, writing text is an extremely painful process. It requires frankness and honesty, a lot of patience, and intellectual effort. But even when you want or need to write, you cannot. Looking at a blank slate
-
bytes) and the constant influx of new digital content poses a real challenge for archivists. Hard disks are expensive and require a constant supply of electricity, while even the best 'no-power' archiving materials such as magnetic tape degrade within a
-
The atmospheric concentration of ozone-depleting CFCs and halons has been falling since the adoption of the Montreal Protocol in 1987. But even as these chemicals have been phased out of aerosol cans, refrigerators, fire
-
weight.⏎- Would have a body weight that is 15% below than what is considered normal for age and height. They would refuse to maintain it such even after advice⏎- Be obsessed with body image and shape. They often fail to
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: what is the real meaning of it? it is a fact, but it's not sure that it is the right? but why, but why it's a fact, but why it's a fact? it is not sure? but why
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Struggle-For-Power-In-The-994323.html>.⏎"The Struggle for Power in the Ussr." StudyMode.com. 0<|endoftext|>Afghanistan and Colombia, half a world apart and different in culture
-
|The Physical Object|⏎|Number of Pages||136|⏎Learn why the Common Core is
-
of Principles of Biblical Interpretation found in the catalog.⏎April 1, 2003⏎by Baker Academic⏎Written in English⏎|The Physical Object|⏎|Number of Pages||180|⏎Hermeneutics gives
-
this past summer.⏎Complete information about survivors could not be confirmed<|endoftext|>A simple example of HTML is:⏎<HTML> <HEAD> <TITLE>This is the Title</TITLE> </HEAD> <BODY
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
- L4:661 excites
- L4:1949 suppresses
- L4:3039 excites
- L10:652 weakly suppresses
- L11:1205 weakly suppresses
- L15:238 weakly excites
- L15:651 weakly suppresses
- L10:2876 weakly excites
- L14:1940 weakly excites
- L15:1095 weakly suppresses
- L9:880 weakly suppresses
- L14:2543 weakly suppresses
- L12:602 weakly excites
- L6:275 weakly excites
- L2:2839 weakly excites
- L4:1209 weakly suppresses
- L15:451 weakly excites
- L6:2217 weakly excites
- L11:2273 weakly suppresses
- L15:740 weakly excites
- L14:1930 weakly suppresses
- L13:11 weakly suppresses
- L11:1213 weakly excites
- L13:2520 weakly excites
Feeds later layers
- L18:1130 excites
- L17:1995 excites
- L20:458 excites
- L19:1590 excites
- L18:1915 excites
- L17:39 weakly excites
- L24:148 weakly excites
- L24:116 weakly suppresses
- L27:3043 weakly suppresses
- L19:832 weakly suppresses
- L18:598 weakly suppresses
- L22:1633 weakly suppresses
- L24:2464 weakly excites
- L22:1067 weakly suppresses
- L19:1793 weakly suppresses
- L17:1596 weakly excites
- L26:1661 weakly suppresses
- L22:571 weakly suppresses
- L19:800 weakly suppresses
- L17:2902 weakly suppresses
- L17:2801 weakly suppresses
- L18:2138 weakly suppresses
- L17:1027 weakly suppresses
- L19:1689 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
- unge
- &P
- éré
- 在过渡
- ��
- ighet
- prs
- aced
- >Password
- isseur
Suppresses
- 又
- IELD
- arious
- course
- orrh
- 又称
- often
- щин
- 有多大
- ines