"in" after a superlative or ordinal
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 "in" when it follows a superlative or "second," "only" or "last" noun, as in "the second in a series" or "the only life in the Universe." It is quiet on sentence-opening "In," which triggers its negative side. Turning it up pushed the model into repeated group and idea phrases.
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.
-
ately I have been reading a biography on John D. Rockefeller and wanted to share some of my insights.⏎John D. Rockefeller was one of the most successful businessmen in American history owning 1.5% of U.S. GDP in 1
-
Reference Count: N/A⏎Hand in Hand. Skills Activities for Tutors and Students.⏎Biles, James J.; And Others⏎The guide is the second in a two-volume series designed to provide volunteer tutors of English as a Second Language (
-
22 November 2021 | Vibhav peri | Crises⏎The Great Slaty Woodpecker, once the largest woodpecker in India, is now the largest woodpecker in the world.⏎This is because
-
scientists have made a surprising discovery: a gigantic aquifer of relatively fresh water trapped in porous sediments lying below the salty ocean.⏎A new book, the second in a series of primers with the Earth Institute imprint, provides an interdisciplinary overview drought
-
biology and is the longest-running citizen survey in the world.⏎The Matagorda County Mad Island Marsh Christmas Bird Count (CBC) is one of the most successful in the country every year and is typically the reigning champion when it comes to number of
-
continental unease between South America and Europe.⏎It led to several big football nations missing out on the tournament.⏎The third World Cup turned out to be the last in the pre-World War II era. The world had to wait 12
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 don't know what time of the year I want to have my own life. I think I have to choose my own life, which I have been in the world, since the other time of the year, which I have been.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, messages, and information like recipes or instructions. In many cases, content like photos or videos is shared. Read on to hear more
-
, it not only affects the productivity of teams, but it can also potentially hurt relationships with customers or damage the reputation of the business.⏎How dirty data happens⏎In Salesforce and most other CRMs, the origin of dirty data can usually be pinpoint
-
changed between Commit 1 and Commit 3.⏎The diff will compare the differences between the two commits, showing what files were added, deleted, or modified.⏎In this example, Commit 1 is the baseline. You may also notice how the
-
wastewater lagoons across the state.⏎It would fail to remove harmful bacteria that impair streams and cause human illnesses, including dysentery and hepatitis A. In that case, even streams that gained new protection wouldn't see an improvement in water quality
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
- L4:289 strongly excites
- L10:262 suppresses
- L14:1389 suppresses
- L8:351 suppresses 'in' after 'high' or 'dressed'
- L9:1025 excites
- L15:871 suppresses
- L3:493 suppresses
- L1:1179 suppresses 'in' after 'result' (results in)
- L14:1625 excites the word 'in' after a noun
- L5:490 excites
- L6:3070 suppresses
- L3:311 suppresses
- L4:1966 suppresses
- L7:2192 weakly excites
- L15:1328 weakly suppresses
- L1:1922 weakly suppresses
- L2:2307 weakly excites
- L11:1781 weakly excites
- L15:1468 weakly excites
- L5:729 weakly suppresses
- L4:1406 weakly excites
- L15:1642 weakly excites
- L3:1116 weakly excites
- L9:2862 weakly suppresses
Feeds later layers
- L17:2877 suppresses
- L19:1354 suppresses
- L23:967 suppresses
- L21:1935 suppresses
- L17:2126 weakly suppresses
- L17:505 weakly excites
- L19:1910 weakly suppresses
- L18:2904 weakly excites
- L18:2012 weakly suppresses
- L17:2314 weakly suppresses
- L25:2083 weakly excites
- L17:1574 weakly suppresses
- L24:893 weakly excites
- L20:2692 weakly excites
- L18:726 weakly excites
- L23:465 weakly suppresses
- L24:2584 weakly suppresses 'in' in set phrases like 'in mind'
- L25:2459 weakly excites
- L22:938 weakly suppresses
- L17:1562 weakly excites
- L17:718 weakly excites
- L17:2114 weakly excites
- L17:3069 weakly excites
- L19:1733 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
- aviours
- facts
- 夤
- Aggregate
- уществ
- lies
- idot
- aggregates
- Reads
- angement
Suppresses
- 没有必要
- atorial
- 收到了
- oping
- xic
- =view
- 没必要
- 直播间
- vh
- zew