"is" after "it" in advice phrasing
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 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 fires on the verb right after a dummy "it" in advice sentences, such as "it is important," "it is best" or "it pays to." It is usually the word just before a judgement about what to do. Turning it up gave mostly unrelated text, so the steering doesn't add much.
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.
-
Higher performance video cards provide smoother full motion video and better 3D gaming performance.⏎How Video Cards Operate⏎Before selecting a video card, we believe it is important to know what a video card does in the system. The separate video card
-
a unique continental and leading agricultural product. One of the countries on the equator line benefits from its climate and agriculture production. Before you go further, it is good for you to enrich your agriculture production countries here. Now, let’s get
-
You can communicate a thought, but your thought may not be understood. In some cases, your thought may not even reach the proper target. Thats why it pays to ask questions to make certain that people understand what you are saying. The great
-
the appSettings element and other configuration sections that a developer might define. It then looks in the application configuration file. To keep the machine configuration file manageable, it is best to put these settings in the application configuration file. However, putting the settings
-
point so that different grouped assemblies could be aligned according to their alignment points, thus aiding in mating parts together.⏎Third: Align to Center or Outside! It would be nice to align the center or outside edge of one object to the edge of another
-
potential that the technologies suggest. Potential that deserves to be explored and brought to market.⏎Before addressing the potential impact of the research and the responses to it, it makes sense 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: I am going to be part of a group. I have been going through a phase of decision-making, and I am going through a phase of decision-making. I have been going through a phase of decision-making, and I have been going
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
1991. As Dr. Wade Huntley puts it:⏎According to this view, had it not been for a tendency toward wild-eyed anti-communism on the American side, the
-
, and I will be your God, and you will be My people; and you will walk in all the way which I command you, that it may be well with you.'⏎Matthew 7:24-27 “Therefore everyone who
-
students to shine through, I feel like I may have caught a glimpse of that, yet it was fleeting. My next steps are to reflect on what it was about the CSI routine that brought that thinking out.<|endoftext|>GNU Octave is a high
-
.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce:v:42:y:2011:i:2
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
- L5:1885 strongly excites
- L5:1807 suppresses
- L4:636 suppresses
- L9:2252 suppresses
- L6:572 suppresses
- L9:1243 suppresses
- L9:1907 excites
- L6:2010 excites
- L7:1833 suppresses
- L5:478 weakly suppresses
- L6:1922 weakly excites
- L7:1970 weakly suppresses
- L7:2106 weakly suppresses
- L7:31 weakly suppresses
- L6:13 weakly excites
- L6:1829 weakly suppresses
- L7:1469 weakly excites "(" before a year in citations
- L4:2583 weakly suppresses
- L8:10 weakly excites
- L2:2648 weakly suppresses
- L4:1320 weakly excites
- L2:2193 weakly excites
- L9:2056 weakly excites
- L7:1163 weakly suppresses
Feeds later layers
- L17:1773 excites
- L13:645 excites
- L12:1022 suppresses
- L13:2186 suppresses
- L14:534 suppresses
- L11:1046 suppresses
- L11:2158 excites
- L12:850 suppresses
- L18:2546 excites
- L13:921 excites
- L12:525 excites
- L13:720 suppresses
- L13:1228 excites
- L14:1309 excites
- L11:2956 suppresses
- L16:2102 excites
- L15:1973 suppresses
- L17:503 excites
- L15:2636 weakly suppresses
- L15:1092 weakly excites
- L16:1100 weakly excites
- L18:727 weakly suppresses
- L14:1062 weakly suppresses
- L16:1985 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
- fact
- łość
- окумент
- learn
- WARE
- empo
- ách
- 釜
- 隆
- ruc
Suppresses
- 捺
- 主要包括
- positions
- coverage
- ,default
- الداخلية
- atable
- бу
- itself
- ittance