the '0' in 360
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. 3 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 to the number 360, specifically the last digit, as in '360 camera' or '360 degrees'. Turning it up did not bring out this theme in the model's replies.
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.
-
look, just as I am sure you can picture a bar code in your mind just by my mentioning them.⏎I created a video that transitions through 360 QR Codes so you can begin to see what changes, and what stays the same
-
converts rotary motion into linear motion.⏎The ‘lead’ of a Lead Screw is the distance which a Lead Screw Nut will move during one 360° rotation of the screw. A smaller Lead results in a tighter thread, which
-
. Ramnarine has also taught architecture students how to use 360 technology to visualize projects.⏎“We showed them how to take 360 photos
-
Luksatrio is using to create his virtual tour project. Once it’s complete, he’ll create a manual for future students to use 360 cameras and begin introducing the technology into their program.⏎To learn more about 3
-
projects more engaging. By capturing videos and photos from every angle, it allows viewers to move and interact with the images.⏎The creative possibilities of a 360-degree camera prompted second-year professional communications student, Christo Luksatrio,
-
] when you’re not not restricted to a single frame.”⏎The department of professional communications has purchased a 360fly, a user-friendly 360 camera, which
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 have been a detective story. I have been a criminal. I have been a recluse. I have been a recluse. I have been a reclusive hero. I have been a reclusive hero. I have been a re
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
—Using narrow, intermediate, or broad definitions, the probandwise concordance for alcoholism was consistently higher in monozygotic than in dizygotic twin pairs. Multifactorial threshold models suggested that the heritability of
-
are some of the classes in which we should expect and reward failure: if our students don’t make mistakes, they aren’t trying anything new.⏎Science teacher Marsenne Kendall tells parents at Back-To-School night: “"I embrace mistakes!!
-
life like on the ice? Our team of six guys and two girls flew to Christchurch, New Zealand, and then were flown down to the US McMurdo base in Antarctica. We spent a week
-
ume is streaming southward from the eastern shore of Pensacola Bay.<|endoftext|>Ticks are frightening pests. In fact, these eight-legged and hard-bodied arachnids can scare even the most hardened horse owner. They are small and difficult 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
- L2:2317 excites
- L0:1224 weakly suppresses
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L2:1312 weakly suppresses
- L2:3061 weakly excites
- L1:189 weakly suppresses
- L2:49 weakly suppresses
- L2:2328 weakly suppresses
- L1:951 weakly excites
- L1:2922 weakly excites
- L0:2709 weakly excites
- L1:331 weakly suppresses
- L1:112 weakly excites
- L2:1710 weakly suppresses
- L2:2256 weakly suppresses
- L0:248 weakly suppresses
- L1:247 weakly excites
- L2:1928 weakly excites
- L2:1262 weakly suppresses
- L1:777 weakly excites
- L2:2947 weakly suppresses
- L2:2221 weakly suppresses
- L0:858 weakly excites
Feeds later layers
- L4:1947 excites
- L7:1433 weakly excites
- L22:1574 weakly excites
- L21:1129 weakly excites
- L4:2310 weakly suppresses
- L10:879 weakly suppresses
- L5:2450 weakly excites
- L4:2104 weakly suppresses
- L9:1482 weakly excites
- L5:2236 weakly suppresses
- L20:1959 weakly excites
- L13:298 weakly excites
- L21:2608 weakly excites
- L19:1908 weakly excites
- L22:422 weakly excites
- L6:1358 weakly excites
- L4:1452 weakly excites
- L4:2483 weakly suppresses
- L19:1976 weakly excites
- L7:2334 weakly excites
- L9:2704 weakly excites
- L10:1226 weakly suppresses
- L26:2906 weakly excites
- L23:256 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
- 圜
- ='_
- equalTo
- imens
- =".
- LinkId
- Ey
- =\"/
- Tac
- TestId
Suppresses
- 脑
- ور
- 之
- ven
- will
- 走到
- tty
- すれば
- 必
- 辈