"by" after a passive verb
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. 13 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 word "by" when it names who made or did something, as in "created by", "written by" or "betrayed by". Turning it up gave wandering text that did not clearly show the pattern, so the effect is hard to read.
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.
-
Megan Smith0⏎A Hummingbird’s Amazing Journey⏎by Robert Burleigh⏎- Language: english⏎- Format: hardcover,
-
in italics) are related in the sentences to follow. A<|endoftext|>southeast asia 2⏎Home > Preview⏎The flashcards below were created by user⏎on FreezingBlue Flashcards.⏎by early 1900
-
and several regiments of infantry, while the heights on both sides were crowned by Russian artillery and infantry. Lord Raglan, perceiving that it was the intention of the Russians to carry off the guns they had captured from the Turks, ordered up
-
|⏎Koothus in Silapathikaram⏎by Subashini Pathmanathan⏎Ancient Tamils called their dances by the
-
just that. After discussion with another friend, Wes Prichard, Prichard suggested to Wennberg<|endoftext|>Home > Preview⏎The flashcards below were created by user⏎on FreezingBlue Flashcards.⏎What factors could affect feet and leg
-
and found a hiding place for them for the remainder of the war.⏎Waterman Pinkhof continued her work sleeping at a different address each night until she was betrayed by someone claiming to be able to assist in Joop Westerweel’s release
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: a question from a guest of a cat. all, a cat. there were three, a white fluffy white, fluffy, white, white., the white fluffy white, white, white, white, white, white, three
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, and others reinforcing existing knowledge and challenges. But all have important implications for designing appropriate programme interventions to fill information and communication gaps. They include:⏎- The dominant role of radio, particularly community radio, as an information medium and potentially a
-
may have taken, few of them have endured, which helps explain why they've been so easily and so largely forgotten.⏎On earlier occasions, too, the part played by Greece has been pivotal - sometimes positive but sometimes negative. And history has recently
-
,” said Tom Crouch, a senior curator at the museum and the man in charge of coordinating the event.⏎“It is a chance to remind visitors of the role that the Smithsonian has played in the history of flight,” Crouch said, “from
-
the historical conception of the county as the place of vicinage in federal capital trials; (2) a Batson-type three-step process for rooting out the influence of race on the decision to prosecute federally; and/or (3) voluntary measures
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
- L12:1286 strongly suppresses
- L9:1562 excites 'by' before a byline or agent
- L12:2810 excites
- L12:1529 excites
- L12:321 suppresses
- L11:2517 excites
- L6:2880 suppresses
- L12:337 suppresses
- L10:2138 excites
- L12:2016 suppresses
- L12:1040 suppresses
- L12:1531 excites
- L12:2266 suppresses
- L11:2387 suppresses
- L10:1054 suppresses 'from' after movement or transfer verbs
- L11:1450 suppresses
- L12:1066 suppresses
- L12:1304 suppresses
- L9:1465 excites
- L9:355 suppresses
- L6:1637 excites
- L12:2369 weakly suppresses
- L3:2203 weakly suppresses
- L11:2497 weakly excites
Feeds later layers
- L14:2374 suppresses
- L20:1357 suppresses
- L14:1711 suppresses
- L18:2195 suppresses
- L16:630 suppresses
- L18:2321 suppresses
- L15:3006 suppresses
- L19:1832 suppresses
- L14:2918 suppresses
- L17:1033 suppresses
- L16:2530 suppresses
- L16:1706 weakly suppresses
- L14:1796 weakly suppresses
- L19:1503 weakly suppresses
- L17:1370 weakly excites
- L17:1221 weakly suppresses
- L15:2017 weakly suppresses
- L14:2007 weakly suppresses
- L14:1953 weakly suppresses
- L16:1986 weakly excites
- L19:2617 weakly suppresses
- L17:577 weakly excites
- L17:1058 weakly suppresses
- L18:2770 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
- 渥
- станавлива
- ynthesis
- >()->
- networking
- 后台
- Networking
- 在网络
- mix
- 在京
Suppresses
- emple
- 这名
- ôm
- donor
- unicorn
- spacer
- Pence
- 瓒
- óm
- Contributors