colon in clock times
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. 23 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 lights up on the colon in times of day, like the colon in '10:00' or '6:37'. Turning it up made the model write in a loose, repetitive way, with no clear topic.
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.
-
Moon will be full on Tuesday night, April 7, 2020, appearing opposite the Sun (in Earth-based longitude) at 10:
-
local wind shear was having its effect on the storm, and Hurricane Humberto’s winds had begun to diminish. By 1500 UTC (11:00 a.m), Humberto had weakened to tropical storm status, with maximum
-
before making an official announcement.⏎That announcement has just come. Gravitational waves were observed on 14 September 2015, at 5:51am ET by both of the LIGO detectors, located in Livingston,
-
mission," ESA Director-General Jean-Jacques Dordain declared at the International Astronautical Congress in Glasgow, Scotland.⏎The end came at around 9:30 a.m. ET, when controllers back at Europe's mission control in
-
the Pacific theater of WWII when she attacked and sank a midget submarine, perhaps the same one, at 06:37. Five midget submarines had been assigned to torpedo U.S. ships
-
dream up the basic elements of the classroom of the future using design thinking. Great example of a school-business partnership.⏎- Design & Thinking Official Trailer (02:28)⏎Design & Thinking is a documentary released this year about design thinking.
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: can we can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can can
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
information about how to correct material in RePEc.⏎For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: (Michael McNulty)⏎If references are entirely missing, you can
-
dates and are celebrated in all Afghanistan.⏎- Juma (Friday):⏎The week ends on Friday (Juma) which is the day of rest.⏎- Jeshen (Independence day): 28th Asad = 19
-
ːa]) was one of the Three Kingdoms of Korea, and one of the world's longest sustained dynasties. Although it was founded by King Park Hyeokgeose, the dynasty was ruled by the Gyeongju Kim
-
• Encyclopedia of Alternative Energy • Contact<|endoftext|>Map of Liberia, West Africa⏎In 1815, the African-American Quaker and maritime entrepreneur Paul Cuffee financed and captained a voyage to Sierra Leone, where he helped a small
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
- L20:1691 suppresses
- L21:1045 weakly suppresses
- L7:92 weakly excites
- L19:2465 weakly excites
- L20:261 weakly suppresses
- L3:828 weakly suppresses
- L20:1732 weakly suppresses
- L9:2831 weakly excites
- L19:945 weakly suppresses
- L17:1493 weakly suppresses
- L19:2180 weakly suppresses
- L11:1603 weakly suppresses
- L18:2382 weakly excites
- L18:1187 weakly excites
- L16:2849 weakly suppresses
- L13:722 weakly excites
- L19:984 weakly excites
- L18:714 weakly suppresses
- L13:1455 weakly suppresses
- L4:42 weakly suppresses
- L19:1648 weakly excites
- L0:1914 weakly suppresses
- L19:730 weakly excites
- L9:2232 weakly excites
Feeds later layers
- L27:774 strongly suppresses
- L24:766 suppresses
- L24:789 suppresses
- L27:1815 suppresses
- L24:800 excites colon in clock times
- L26:2879 weakly excites
- L27:2857 weakly suppresses
- L24:1298 weakly excites
- L27:2782 weakly suppresses
- L24:1324 weakly excites
- L24:1814 weakly excites
- L24:797 weakly excites
- L26:1401 weakly suppresses
- L27:1592 weakly excites
- L27:1699 weakly suppresses
- L25:1952 weakly suppresses
- L25:2844 weakly excites
- L24:1183 weakly excites
- L27:2378 weakly excites
- L24:2964 weakly excites
- L24:2367 weakly excites
- L27:2960 weakly excites
- L24:2804 weakly excites
- L24:1740 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
- }}},⏎
- ugh
- 他会
- oundation
- acet
- ッド
- 管局
- enen
- ugar
- 打法
Suppresses
- 可
- 可以
- 可在
- can
- 可以在
- can
- 可以从
- Can
- _can
- Can