numbers and dates with stray punctuation
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. 0 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 reacts to small punctuation pieces in numbers and dates, like the dot in "2.M" or the dash in "1910-Moved". It is a weak signal, and turning it up gave text that did not show any clear 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.
-
2 students learn to tell and write time from analog and digital clocks to the nearest five minutes, using a.m. and p.m.⏎Common Core: 2.MD.7⏎
-
of Pharmacy terminated.⏎1905-Theological Seminary organized in Waco.⏎1907-Separated from Baylor University.⏎1910-Moved to Fort Worth.⏎1918-College of Dentistry organized in
-
along with other associations, sent representatives in 1848 to create the Baptist State Association, which later became the Baptist State Convention.⏎1845-Baylor chartered on February 1 by the Republic of Texas.⏎184
-
.⏎1996 -The College became a part of the Texas A&M System on September 1, 1996.⏎1919-Baylor Hospital organized in Dallas – now Baylor University Medical Center.⏎1919
-
_304 Measurement System Inch Number of Items 1 Part Number 190.0400.0400.025.SS UNSPSC Code 30260000 See more
-
80E.2s in 1919, joining seven Nieuport 83.E.2s that they had purchased in 1916.⏎In
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 thinking about how to get a good score in a game. And I have been thinking about how to get a good score in a game. I have been thinking about how to get a good score in a game. And I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
North American continent.⏎Rising prominently at 5,895 m (19, 341 ft) in north-eastern Tanzania, the snow-capped
-
72,000 km/hour (45,000 miles/hour) when it hit the atmosphere. As an asteroid
-
m (18,510 ft) on its western summit and 5,621
-
Curtiss and purchase a plane for $7,000 ($157,000 in today’s money). While waiting for it to be built,
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
None: this is the first layer.
Feeds later layers
- L4:2598 suppresses
- L8:592 weakly suppresses
- L6:2063 weakly excites
- L7:1654 weakly excites 'memory' in brain and ageing text
- L26:385 weakly suppresses
- L6:1240 weakly excites
- L2:1039 weakly excites
- L19:0 weakly excites
- L5:365 weakly suppresses
- L6:697 weakly excites
- L6:58 weakly suppresses
- L3:2392 weakly excites
- L4:779 weakly excites
- L16:1218 weakly suppresses
- L7:2614 weakly excites
- L20:628 weakly excites
- L7:2421 weakly excites
- L22:2385 weakly excites
- L25:112 weakly suppresses
- L6:182 weakly excites
- L3:732 weakly excites
- L22:1316 weakly excites
- L9:906 weakly excites
- L10:780 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
- amus
- ró
- elow
- ATA
- rush
- !
- !
- ingroup
- -pro
- ở
Suppresses
- 杀手
- 利用率
- 计算器
- 打击
- tester
- 鸡肉
- thon
- setSupportActionBar
- banker
- eteria