the word 'even' as emphasis
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. 12 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 word 'even' when it adds emphasis or a surprise, as in 'even the most pessimistic' or 'don't even think about'. It is quiet on most other words. Turned up, the model's text wandered without a clear link to this word.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
timeframe⏎- Low: Prioritize and decide whether to<|endoftext|>Stunning new data not yet publicly released shows Louisiana losing its battle with rising seas much more quickly than even the most pessimistic studies have predicted to date.⏎While state officials continue to argue
-
days by a fraction, hardly had an impact on the planet in the long run. In fact, scientists have a hard time spotting the effect of even bigger quakes on something such as the Earth's rotation, said Richard Gross,
-
the Titanic, less than half survived. There were only 20 lifeboats aboard the White Star Line luxury ship, enough to carry half those onboard, but even these were not filled to capacity. When the liner Carpathia reached the disaster
-
cover stuff that well,” says Rapini, who was not involved in the research.⏎“Myself, personally, I’m a dermatologist, and I don’t even wear sunscreen all that much. I stay in the shade,” he says. “
-
9/11s every week for the last 7 years⏎, would you know which one I'm talking about?⏎I bet few westerners could, even though it's by far the deadiest conflict of the last 60
-
one’s mind around the concept of black holes. One teaspoon can be estimated to have tons and tons of weight. They are so dense that many times light cannot even escape. or at least that is what astronomers think. Are they right?⏎
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 use of the “S” in the name of the car? So, why is the name "Car" relevant? What is the “S” in the name? What is the “S” in the name? I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
exact cooking times for<|endoftext|>Here are the 15 types of distorted thinking:⏎1. Filtering: You take the negative details and magnify them, while filtering out all positive aspects of a situation. A single detail may be picked out, and
-
turn toward a more populist brand of politics has more to do with the energy of reform movements within Honduras itself, as well as throughout the rest of Latin America, than any personal awakening.⏎From the shift to civilian rule in 19
-
or policies of the West - chiefly the United States - were actually not very important. He implicitly affirms the idea that the Soviet Union fell to domestic factors, dismissing the idea that an external power could cause such domestic upheaval as "childish
-
causing the periphery to gradually fade away, not only will the photo be more like actual vision, it can immediately draw the eye to the main subject and away from unimportant elements in the background or periphery. It
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
- L11:2083 suppresses
- L11:1095 suppresses
- L9:2535 suppresses
- L11:2937 excites
- L7:32 excites
- L9:1899 weakly suppresses
- L8:1962 weakly suppresses
- L4:151 weakly excites
- L9:54 weakly excites
- L7:662 weakly excites
- L8:2348 weakly excites
- L5:6 weakly suppresses
- L11:1006 weakly excites
- L5:45 weakly excites
- L6:1190 weakly excites
- L6:1175 weakly excites
- L10:2069 weakly excites
- L5:2058 weakly excites
- L11:2540 weakly excites
- L7:650 weakly excites
- L6:1477 weakly suppresses
- L4:9 weakly excites
- L0:653 weakly suppresses
- L4:2802 weakly excites
Feeds later layers
- L14:745 suppresses
- L20:952 suppresses
- L19:2606 suppresses
- L16:2401 suppresses
- L15:2954 suppresses
- L14:480 suppresses
- L13:2356 suppresses
- L18:825 suppresses 'as well as' and list connectors
- L14:1217 suppresses
- L14:2166 suppresses
- L16:1735 suppresses
- L14:651 suppresses
- L14:2882 suppresses
- L15:2455 suppresses
- L17:1539 suppresses
- L14:1859 weakly suppresses
- L16:686 weakly suppresses
- L16:964 weakly suppresses
- L16:2591 weakly suppresses
- L16:1282 weakly suppresses
- L17:1017 weakly suppresses
- L26:622 weakly excites
- L15:2205 weakly suppresses
- L15:2226 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
- lesh
- INESS
- 委组织部
- �
- сла
- scriber
- ishops
- 部副部长
- ович
- ицы
Suppresses
- 其他
- other
- 其它
- other
- 其余
- 其他的
- Other
- 其他国家
- andere
- Other