'all' in 'at all' phrases
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. 4 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 responds most to the word 'all' in phrases like 'not at all' or 'at all'. Its peaks are fairly weak compared with the strongest neurons. Turning it up or down did not clearly change the topic of the output.
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.
-
he disliked the Panadavas. He also knew that Krishna always favoured the Pandavas, because they were always on the side of righteousness. If at all there was a war between
-
makes you wonder why Moses included it where he did, or at all. And even more, "Why, now is this seemingly obscure story from the
-
in their discussion of creating an entrepreneurial mindset.⏎Because entrepreneurs bring together the factors of production in order to make a profit, I actually don't think entrepreneurs are at all what we need.⏎I've heard Zhao mention that students, if they want to
-
this:⏎- Sandy's wearing a lot of makeup, but it's not at all whorish. It's actually rather well done.⏎Unfortunately, while you know
-
019, The World Health Organization issued its first-ever guidance for how much screen time children under 5 should get: not very much, and none at all for those under 1.⏎The U.N. health agency said that kids under
-
the highway, Phil finds himself trapped in Punxsutawney. He wakes up the next day, only to discover that it's not the next day at all. It's Groundhog Day all over again.⏎via Wikimedia Commons⏎For some
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: to have a more concrete answer. I need to make sure that I can get it. So, the best way is to get a good deal. So, the user is clear. But I don't know. Maybe I need to have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
number follows it. From UPF10 to UPF50+, each number offers a certain level of protection.It’s simple: the higher the number, the more protection you get. UPF50, for example, means that only
-
help. The sooner you receive care for your anxiety symptoms, the better. Feel free to reach out to Neuro-Behavioral Clinical Research for help
-
have a heavy clay soil. Apply less water less frequently.⏎Water 10-15 gallons per week with a slow drip. The heavier the soil, the less water it may need. Watch to see how fast the water drains and avoid
-
on average. Diastolic blood pressure (the bottom number) decreased by 6 to 7 mmHg, on average. The higher the dose, the greater the drop in blood pressure tended to be. By lowering blood pressure, Mon
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
- L1:2470 weakly suppresses
- L2:111 weakly suppresses
- L2:1411 weakly suppresses
- L2:1076 weakly excites the 'a' after 'not' or 'just'
- L0:450 weakly suppresses
- L0:633 weakly excites
- L3:421 weakly suppresses
- L1:2827 weakly suppresses
- L3:1937 weakly excites
- L2:944 weakly suppresses
- L3:552 weakly suppresses
- L0:3070 weakly excites
- L1:1287 weakly excites
- L1:1722 weakly suppresses
- L3:456 weakly suppresses
- L2:971 weakly excites
- L3:2322 weakly suppresses
- L1:1035 weakly suppresses
- L3:2337 weakly suppresses
- L0:847 weakly suppresses
- L3:2013 weakly excites
- L1:303 weakly suppresses
- L1:2021 weakly excites
- L3:1768 weakly excites
Feeds later layers
- L7:747 suppresses
- L18:1128 weakly excites
- L15:2896 weakly excites
- L5:1701 weakly suppresses
- L8:1960 weakly suppresses
- L5:2850 weakly suppresses
- L5:2316 weakly suppresses
- L7:2495 weakly excites
- L11:2190 weakly suppresses
- L7:1277 weakly suppresses
- L6:1681 weakly suppresses
- L25:545 weakly suppresses
- L23:2875 weakly excites
- L17:199 weakly suppresses
- L17:1205 weakly suppresses
- L5:2587 weakly suppresses
- L18:2890 weakly suppresses
- L20:729 weakly excites
- L20:1197 weakly excites 'this' pointing at the current document
- L6:885 weakly suppresses
- L11:1522 weakly excites
- L14:660 weakly suppresses
- L17:1462 weakly suppresses
- L19:2903 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
- 加工厂
- ception
- junction
- starter
- .jquery
- adow
- ctype
- gle
- ál
- ieves
Suppresses
- 如意
- abdomen
- 遏制
- 督办
- swear
- compact
- Advent
- 不满意
- Compact
- Endian