the word 'late' in time phrases
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. 16 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 fires on the word 'late' when it is used to point at a period of time, like 'the late 19th century', 'in late 2009' or 'late summer'. Its negative side picks up odd spots inside words in other scripts. Turning it up or down produced text that did not show any clear theme.
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.
-
derived from living matter were applied to wounds to destroy infection, the fact that a microorganism is capable of destroying one of another species was not established until the latter half of the 19th cent. when Pasteur noted the antagonistic
-
Boston, he was a declared music major. Late in his collegiate career, he discovered a love of biology – particularly the brain and
-
combination of structural weaknesses of the Greek economy along with a decade long pre-existence of overly high structural deficits and debt-to-GDP levels on public accounts. In late 2009, fears of a sovereign debt crisis developed among investors concerning
-
000 acres of pastures for cattle and sheep that, starting in the late 19th century, gave way to development of the cities of Long Beach
-
dots to the naked eye. Despite their size, our arborists warn that mites can be a huge problem for homeowners, especially in the hot days of late summer.⏎When temperatures are over 80 degrees, oak spider mites reproduce
-
especially the United States, which retained ties with Chiang in Taiwan, recognising his as the legal Chinese government.⏎Since embarking on landmark reforms beginning in the late 1970s, the country has emerged from isolation to become the
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 number of people who have been in a place of power, or control over control over control, in the world. That’s a huge number, not just in the US or the United States. We can't understand how that's been
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
' is from the words 'Gur (ਗੁਰ)' (a reference to the Sikh Gurus) and 'Dwara (ਦੁਆਰਾ)' (gateway in Gurmukhi), together meaning
-
The Court of Appeals of Alabama reversed the conviction, and the Supreme Court of Alabama decided that the licensing provisions were constitutional and reversed the appellate court. The Jehovah’s Witnesses from Alabama appealed to the U.S. Supreme Court, which consolidated the
-
recognizably in the Hebrew name of The Jewish Agency – הַסּוֹכְנוּת הַיְּהוּדִית, or simply הסוכ
-
of nose bleeds. Because the membrane and blood vessels inside the nose are so sensitive, any irritation caused by the picking can lead to bleeding. Frequent, forceful nose-blowing can also be a factor. People who are prone 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
- L15:1587 strongly excites
- L15:1900 strongly suppresses
- L15:1404 strongly suppresses
- L15:792 strongly suppresses
- L15:381 strongly excites
- L15:222 strongly excites
- L15:1135 strongly suppresses
- L15:1616 strongly excites word fragments mid-word
- L15:435 strongly excites
- L15:201 strongly excites
- L15:451 strongly excites
- L15:2803 strongly suppresses
- L15:1164 strongly excites
- L15:787 strongly excites
- L15:12 suppresses
- L14:1090 excites
- L15:1300 excites
- L14:126 excites
- L14:169 excites
- L15:966 suppresses
- L15:227 excites
- L15:1043 suppresses
- L15:1394 suppresses
- L15:554 suppresses
Feeds later layers
- L19:2126 excites
- L19:1470 excites
- L27:775 suppresses
- L20:2064 excites
- L18:168 excites
- L19:833 suppresses
- L19:2724 suppresses
- L19:1016 excites
- L26:2022 excites
- L22:470 suppresses
- L26:1374 weakly suppresses
- L26:3023 weakly excites
- L26:1698 weakly excites
- L21:1736 weakly suppresses
- L27:606 weakly suppresses
- L18:3058 weakly excites
- L22:2027 weakly excites
- L26:1015 weakly suppresses
- L22:423 weakly excites
- L20:603 weakly suppresses
- L27:1007 weakly excites
- L20:2097 weakly suppresses
- L26:1861 weakly suppresses
- L26:1545 weakly suppresses
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
- ning
- res
- anced
- stab
- ell
- 목
- oted
- 恙
- haste
- plate
Suppresses
- SBATCH
- sylvania
- orial
- ATIONAL
- '})
- )})
- ска
- )});⏎
- Frm
- stakes