'still' or 'yet' after a contrast
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. 14 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 words like 'still' or 'yet' when the text has just set up a surprise, such as 'even though X, they still Y'. Turning it up made the model's writing drift to unrelated topics rather than anything about contrast.
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.
-
hesives,” holding certain proteins to other RNA molecules or to DNA.⏎Now even though the last two jobs I mentioned are not directly related to protein production, they still involve proteins. So is it safe to say that while RNA performs several functions in
-
implementing mindful eating into your daily life.⏎Have you ever forgotten to eat lunch because you’re so busy throughout the day? Have you ever eaten a meal and then still felt hungry after? Have you ever mindlessly grazed on a snack? Or
-
9 or 10 — they look younger due to malnutrition — and are in grades four or five in their regular government schools. Despite their age, they still don’t know basic math. The toddlers are their siblings, who the girls —
-
Ethnic Minorities<|endoftext|>Lily of the valley, or convallaria, is not a true lily like the Easter or Asiatic lily, but is still toxic to dogs, humans and other animals. If your dog chews on the
-
-thirds of the way from the center to the edge. At 100,000 light years across, the Milky Way is vast beyond comprehension and yet it’s only one of an estimated 100 billion galaxies in the observable
-
for the sake of “looking cooler?”⏎The risk of head injury might be low. But the risk of a car crash is also low, and yet you still wear a seatbelt. The risk of a plane crash is incredibly low, and
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 was just 23, and I was just 23, I was just 23, I was just 23, I was just 23, I was just 23, I was just 2
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
adult before you fill out any forms, or give out money or credit card details.⏎- Don't accept any offers that seem too good to be true - they probably are.⏎Remember to always be
-
have yet to find the perfect thickness, the perfect fill, that they’re certain will reliably support a person’s weight.⏎But with each iteration, they’re getting closer to their goal: a relatively simple, low-cost process for scanning, modeling,
-
scale chemicals makers could be fighting for their very survival if carbon reduction targets produce a particularly hostile operating environment. Cost (of production) is not everything, but it is close to it.⏎That fact is brought home particularly in petrochemicals where
-
murex shellfish, known as Tyrian purple. This color was, in many cultures of ancient times, reserved for the use of royalty, or at least nobility.⏎It was often attacked by Egypt, besieged by Shalman
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
- L12:3057 excites
- L12:2794 suppresses
- L13:3056 weakly excites
- L12:1527 weakly excites
- L9:1755 weakly suppresses
- L13:26 weakly excites
- L13:2169 weakly excites
- L12:2773 weakly excites
- L12:2717 weakly excites
- L11:2712 weakly suppresses
- L13:2930 weakly excites
- L12:1265 weakly suppresses
- L12:1501 weakly suppresses
- L13:1039 weakly suppresses
- L10:2885 weakly excites
- L10:1207 weakly suppresses
- L6:590 weakly excites
- L13:1859 weakly suppresses
- L4:2622 weakly excites
- L11:2178 weakly excites
- L8:2181 weakly excites
- L8:1141 weakly excites
- L8:2081 weakly excites
- L13:1256 weakly suppresses
Feeds later layers
- L15:1462 weakly excites
- L23:2579 weakly suppresses
- L16:915 weakly excites
- L16:696 weakly suppresses
- L17:2773 weakly suppresses
- L19:718 weakly excites
- L27:1397 weakly suppresses "the" in "to the next"
- L17:1552 weakly suppresses
- L22:2536 weakly suppresses
- L27:1593 weakly suppresses
- L15:1331 weakly excites
- L19:2225 weakly suppresses
- L19:212 weakly excites
- L18:1866 weakly suppresses
- L18:44 weakly excites
- L27:2490 weakly suppresses
- L25:2203 weakly suppresses
- L16:599 weakly suppresses
- L23:2110 weakly suppresses
- L26:20 weakly excites
- L22:774 weakly suppresses
- L15:3068 weakly excites
- L18:38 weakly excites
- L17:1739 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
- USE
- av
- because
- 불구하고
- through
- 安全管理
- Mission
- 实
- 仍
- FUNCTION
Suppresses
- ouched
- xhttp
- 毗邻
- .Reverse
- 嗣
- ขยาย
- urge
- imento
- utation
- ingers