unclear: scattered web and commerce text
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. 11 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 weakly on a mix of ordinary words in shopping pages, school material and news, with no clear shared pattern. It is hard to say what it cares about, and turning it up did not reveal a theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
.0400.025.SS UNSPSC Code 30260000 See more.Customers who bought this item also bought.Page 1 of 1 Start over Page 1 of 1 .
-
UP Details⏎Happy Homeschool Month! The whole month of January we will have NEW DEALS every week for you to stock up for the new year! Use code: STOCKUP at checkout to get an EXTRA 10% off sitew
-
a floor prototype of the brand new exploration extravehicular mobility unit (xEMU) in 2019 at NASA Headquarters in Washington. Credit score: Joel Kowsky/NASA⏎One of many principal elements is the cooling garment
-
of this warm-up is for students to reason about two objects that have the same volume but different surface areas within a context.⏎Give students 1 minute of quiet think time followed by a whole-class discussion.⏎Mai’s science teacher told her that
-
control the back and forth movement of the knee.⏎Most injuries to the ACL happen when playing sports, such as football, squash, tennis or skiing (see the box, left). ACL injuries account for 40% of all sports injuries.⏎
-
(in Central America)|⏎|Scientific Name||Lampropeltis triangulum gaigeae|⏎|Conservation Status||Not listed by IUCN|⏎When these snakes are young,
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 number of the last person in the "Bible" of the "Garden of Eden" to the number of the last person in the "Garden of Eden" on a single page. The number of "Bible" of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
forward to coming in every year to see me.⏎Holding space for someone is a big deal, and it doesn’t happen often. Think about it – how many times do you have an encounter with someone and feel their presence is open, attentive
-
Sen. Marshall. “The clock is ticking as parents inch closer to the fall semester and return their children to the classroom. Every parent and physician I’ve talked to has told me if they knew their patient or child’s level of immune response,
-
(through our fingertips) to electronic devices. And it isn’t as if the possibilities of neural interface technologies haven’t been considered, or even capitalized on (just ask Joss Whedon or the Wachowski brothers or Dr. Octopus
-
people and organizations try making sense of things.⏎Have rumors always existed in human history, or are they relatively modern?⏎I think they have always existed. You can find rumors in ancient writing, usually described as something you didn't want to listen to
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
- L10:2960 suppresses
- L7:1562 excites
- L6:2883 excites
- L8:838 suppresses
- L7:2400 weakly excites
- L9:903 weakly suppresses
- L10:1371 weakly suppresses
- L7:2894 weakly excites
- L8:285 weakly excites
- L4:2789 weakly excites
- L8:2122 weakly excites hand and body action verbs in instructions
- L1:2853 weakly suppresses
- L10:3044 weakly suppresses
- L4:2258 weakly suppresses
- L8:2738 weakly excites
- L10:2643 weakly suppresses
- L5:1786 weakly excites
- L8:1609 weakly suppresses
- L8:1818 weakly excites
- L6:391 weakly suppresses
- L10:283 weakly excites
- L10:3056 weakly excites
- L10:1221 weakly excites
- L7:255 weakly excites
Feeds later layers
- L13:1954 suppresses
- L13:395 suppresses
- L15:2000 suppresses
- L14:436 excites
- L13:2319 excites
- L12:1525 excites
- L14:2472 suppresses
- L14:2429 weakly suppresses
- L15:2640 weakly suppresses
- L13:1184 weakly suppresses
- L14:2870 weakly excites
- L13:276 weakly suppresses
- L14:1604 weakly suppresses
- L13:1452 weakly suppresses
- L13:580 weakly suppresses
- L17:2645 weakly suppresses
- L13:2697 weakly suppresses
- L12:681 weakly suppresses
- L13:79 weakly suppresses
- L12:2825 weakly suppresses
- L15:997 weakly excites
- L16:2860 weakly suppresses
- L14:1206 weakly suppresses
- L17:2754 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
- ựa
- othy
- Specialists
- ymm
- elyn
- пока
- nock
- etermin
- endors
- ,+
Suppresses
- scri
- chè
- 侄
- Patri
- ched
- 相近
- 书籍
- ="#"><
- íst
- }`,