digit inside a date or number
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. 5 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 responds to a single digit in the middle of a date or a large number, such as the '2' in 'July 08, 2009' or the zeros in '20,000'. Turning it up made the model repeat itself, so it does not show a clear meaning in 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.
-
Sept. 18, 2001) initiates the Jewish year 5762. In
-
, but the mechanisms underlying these colonization events remain poorly understood (Heyder 1955; Diamond 1986; Luniak et al. 1990; Rasner et al. 2004
-
the only group this can potentially affect. Because of their short life span, the mice in this study reproduced only a few months after chemotherapy, but humans would likely have years in between
-
found itself subject to a migrating swarm of nearly 20,000 honey bees. The species is close to extinction, so the crew on
-
693_000_027⏎It was beginning to get dark, and the softly falling snow was blank
-
January 3, 2013⏎Emissions of mercury and other air toxics from coal-fired
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: Também, Também, \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
wolves.⏎tAsAM ruditashabdena gavAM haMbhAraveNa cha |⏎v
-
dating yields a layer of rocks. Radioactive substances within archaeological. After reading the fossils it contains compared to the most important difference between a system https://pusisister.com/ my class, relative age. Differentiate between relative dating does your neighborhood
-
IFICATION; LEGAL CLUSTERS; GEOGRAPHIC CLUSTERS; TRADE CLUSTERS; ENVIRONMENT
-
the Fogo-Brava⏎platform with 500 m bathymetric contours and 20
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
- L2:39 strongly suppresses
- L3:994 strongly suppresses
- L2:69 strongly excites
- L4:18 strongly excites
- L1:80 strongly excites
- L1:129 strongly excites
- L2:2 strongly suppresses
- L1:397 strongly excites
- L2:60 strongly suppresses
- L2:727 strongly excites
- L3:54 strongly suppresses
- L3:83 excites
- L3:64 suppresses
- L2:76 suppresses
- L4:155 excites
- L3:22 excites
- L4:8 suppresses
- L2:143 excites
- L2:13 suppresses
- L3:280 suppresses
- L2:113 suppresses
- L1:127 suppresses
- L2:281 excites
- L2:145 excites
Feeds later layers
- L6:126 strongly suppresses
- L7:151 strongly suppresses
- L10:35 strongly excites
- L6:22 strongly excites
- L10:895 strongly suppresses
- L7:124 strongly excites
- L13:358 strongly suppresses
- L8:82 strongly suppresses
- L6:303 strongly excites
- L8:139 strongly suppresses
- L6:212 strongly excites
- L6:34 strongly excites
- L11:289 strongly suppresses
- L14:544 strongly suppresses
- L6:374 strongly excites
- L12:661 strongly excites
- L12:1318 strongly suppresses
- L15:2202 strongly suppresses
- L6:109 strongly suppresses
- L14:1590 strongly excites
- L6:115 strongly excites
- L7:197 strongly excites
- L10:350 strongly excites
- L7:27 strongly 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
- rying
- YYS
- thee
- าร
- RY
- 忉
- iam
- ems
- ACY
-
Suppresses
- vanished
- 二氧化
- ocado
- 世界上最
- 了一份
- $http
- 讽刺
- chú
- conj
- 笑了