decade and century numbers like 1950s
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 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 digits of years written as decades or centuries, such as the '9' in '1950s' or the gap before '17th century'. When it was turned up or down, the model's text did not visibly change in a clear way.
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.
-
first time in Italy in 1590. However, until the first decade of the 17TH century, there is no effective evidence of the birth of the telescope
-
970s bought those semi-enclosed pedestal-style pay phones which, for most of us (Superman included) found lacking.⏎By the 1990s there were nearly three million pay phones in America. But now just
-
just wasn’t much technological or scientific progress at that time.⏎What about the first decade of the 20th century? That’s easier. The Zeppelin airship. The
-
positive impacts on Puerto Rico. What major change hit Puerto Rico in the 1990s that would drastically hurt their economy?⏎By the 1950s, the flow of Puerto Ricans to the mainland United States had
-
NATO planning posited early or initial use of nuclear weapons like the 1964 Warsaw Pact plan from the Czech archives. Certainly, by the 1950s, NATO war plans assumed early use of nuclear weapons, even immediate
-
. Although this is a wonderful rifle that It also progressed through World War II, when most of the military relied on old bolt action rifles, by the 1950s the M1 was simply obsolete. By this time,
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 year 1996.50 1985 1945. I 65 1900.00. 60. 1900 1910
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of a Living Trust⏎Sometimes it becomes necessary to petition a court to have a trust terminated. In the State of Texas, Sec. 112.054 of the Texas Property Code governs judicial<|endoftext|>Adjustment to college life
-
explore the Orange Revolution. In this Orange Revolution lesson, students examine the climate of the Ukrainian protest and investigate the role of Canadian government in the protest.⏎3 Views 2 Downloads<|endoftext|>The environment contains those factors ‘surrounding the organization’.
-
-earth wall within the property of Boisbuchet⏎EUR 858: Students⏎EUR 1110: Professionals<|endoftext|>Created on Thursday, 03 July 2014 Written by THE BELLEFONTAINE
-
1700’s in order to win a prize, that the problem was solved.⏎Joan Marie Galat, Pelican ©2012, 978-1-4556-1637-
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
- L6:1469 excites
- L10:868 suppresses
- L10:2461 excites
- L10:220 suppresses hyphen inside compound words
- L7:229 excites first digit of a split number
- L9:13 excites
- L4:973 suppresses
- L6:756 suppresses
- L6:714 excites
- L8:2933 weakly suppresses
- L6:1888 weakly suppresses
- L6:1149 weakly suppresses
- L6:641 weakly suppresses
- L8:2644 weakly excites
- L8:853 weakly excites slot right before a quantity
- L9:438 weakly suppresses
- L5:2113 weakly suppresses
- L2:1668 weakly suppresses
- L2:1582 weakly excites
- L4:53 weakly excites blank slot before a number
- L4:2510 weakly suppresses
- L0:1745 weakly suppresses quantity word after large, high or low
- L10:781 weakly excites
- L8:885 weakly excites
Feeds later layers
- L19:298 suppresses
- L21:175 suppresses
- L16:372 suppresses
- L18:291 excites
- L15:828 suppresses
- L15:2665 weakly suppresses
- L20:95 weakly excites
- L16:1569 weakly suppresses
- L20:382 weakly excites
- L13:1251 weakly excites
- L16:256 weakly suppresses number right after 'age of' or 'ages'
- L14:177 weakly excites
- L25:46 weakly excites
- L19:1428 weakly excites
- L16:434 weakly excites
- L18:991 weakly suppresses
- L12:2853 weakly suppresses
- L18:2897 weakly excites
- L19:2113 weakly excites
- L14:2082 weakly excites
- L25:209 weakly suppresses
- L14:1700 weakly suppresses
- L14:1690 weakly excites
- L19:245 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
- ưu
- avers
- asin
- inds
- "testing
- oke
- VEC
- idUser
- plement
- ası
Suppresses
- 两年
- ties
- 几年
- 两年前
- century
- Hier
- 的做法
- 的高度
- ingham
- 二十年