'something' or 'anything' after a verb
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. 21 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 words like 'something' and 'anything' that come after a verb, as in 'learn something' or 'discovered anything'. When it was turned up, the model kept repeating the word 'important'.
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.
-
to the phone or to other people. Now, people never have to be alone and even when they are with other people, there is still this fear that maybe something
-
- they don't twinkle, they shine," Johnstone says. He spends as much time as possible gazing at the skies: "I haven't discovered anything of importance yet, like maybe a comet or dwarf planet, but I'm still
-
it doesn’t mean that the students from other classes can’t read this article. Of course, they can.⏎You’ll surely learn something from these moral stories at the end.⏎Read some other English moral stories.⏎H
-
. A solar storm would still pose a risk for wandering astronauts, he adds, but there would be time to return to a shielded base.⏎To continue reading this article, subscribe to receive access to all of newscientist.com, including
-
yming with aight in the 1980s. Why doesn’t the OED cite any of them? Have the Oxford boffins got something against DJ Jazzy Jeff & The Fresh Prince? Maybe, but I suspect
-
is crucial for creativity.⏎Do not be afraid to hear others as you compose. There are plenty of different views on writing, and you’re probably going to learn something as you write. Simply listening to what others say may allow you<|endoftext|>It is
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: the 5th and 6th of 7th 8th 9th 10th 11th 12th 12th 12nd 12nd 12nd 1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of on-line web content, which is often made available to customers using the web. This write-up briefly covers the different types of posting as well as goes over just how these different approaches differ from each various other and just how they can be useful
-
Terms Tests are given every six weeks.⏎Many of the literature selections are taken from Norton Anthologies although they can also be found online. Additionally, one or more whole book studies are covered in each course. For the British Literature course, the author
-
ek's father, Sabu, did not approve of their relationship as he knew that Aya's dream was to settle in Alexandria, and he feared that her mixed heritage would corrupt his teachings.⏎Later
-
hind legs, facing towards the light – reaffirming that Pine Martens truly are superstar performers in front of camera traps.⏎The spread of Pine Martens across much of the Scottish mainland north of Edinburgh and Glasgow is a success story for species resilience
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
- L19:2926 suppresses
- L20:1012 excites
- L19:2219 excites
- L19:3017 suppresses
- L19:2122 suppresses
- L18:801 excites
- L20:809 excites
- L18:1596 suppresses
- L18:2398 excites
- L18:482 suppresses
- L5:2687 weakly suppresses
- L14:2609 weakly excites
- L19:2843 weakly suppresses
- L17:2159 weakly suppresses
- L14:1519 weakly excites
- L19:2149 weakly excites
- L5:1410 weakly excites
- L6:1755 weakly suppresses
- L18:2058 weakly suppresses
- L19:2274 weakly excites
- L17:1086 weakly excites 'something' as an object
- L19:914 weakly excites
- L13:2736 weakly excites
- L9:1663 weakly suppresses
Feeds later layers
- L27:2019 excites
- L24:568 excites
- L22:1297 excites
- L24:915 excites
- L23:530 excites
- L23:1323 excites
- L25:1476 excites
- L22:2814 excites
- L23:283 excites
- L25:1891 excites
- L23:1726 excites
- L22:415 suppresses
- L26:1849 excites
- L22:901 weakly excites
- L25:339 weakly excites
- L23:511 weakly excites
- L27:460 weakly excites
- L23:143 weakly suppresses
- L24:372 weakly excites
- L22:691 weakly excites
- L22:2326 weakly excites the word 'well' before -known
- L27:620 weakly excites
- L26:1612 weakly excites
- L23:2291 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
- important
- 重要
- 重要的
- important
- importante
- importantes
- wicht
- важ
- valuable
- 的重要
Suppresses
- imated
- odore
- 其它问题
- ลบ
- olson
- 投资基金
- arella
- 仅代表
- Erect
- 彘