"it" after whether or when
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. 21 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 "it" when it follows "whether" or "when", as in "whether it's" or "when it comes to". Turned up, the model stuck in loops of "it's it it's".
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
but if anyone deserved the description of polymath, it was Cartwright. Born into a landed family in Marnham, Nottinghamshire
-
.⏎Though the results are promising<|endoftext|>In today’s digital era, apps have become an integral part of our lives, catering to our every need and desire. Whether it’s a gaming app or an investing app like Acorn, one cannot underestimate the
-
Abstract: Data science has long considered learning better models and making better decisions based on existing data. With data science systems penetrating into every aspect of daily life, whether it be navigation systems that influence which route you pick for your commute or personalized news delivery
-
readers."⏎—Andrew Piper, McGill University⏎"If there is one book to assign Humanities and Digital Humanities students that provides a wide range of knowledge about text technologies, it is this one. It introduces principal concepts along with ample historical examples, five diverse
-
, and draw in the hands on an analog clock.⏎- I can write the time and draw in the hands on an analog clock when someone tells me what time it is to the nearest
-
to understand the required knowledge that goes into the planning of a house. First Step: The Process: Understanding the fundamentals of the design process is the most important when it comes to designing your own home.⏎Designing a house is the process of shaping
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;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
8211; Basic Chemistrynbsp;nbsp;Chemistry Questions 8211; Learn Basic Chemistry Part 1 (1-25)nbsp;1) When a beam of white rays is dispersed by a prism which colour
-
Children in those communities are at risk of developing severe tooth enamel fluorosis, a condition that can cause tooth enamel loss and pitting. A majority of the report\u0092s authoring committee also concluded that people who drink
-
ner]Associated Press Psychologist Urie Bronfenbrennerdeveloped the ecological model ofhuman development. Urie Bronfenbrenner8217;s ecological model of human development has had essay junk food should be banned enduring impact
-
students across the toronto district school board have been Children who attend preschool enter kindergarten with improved skills but that advantage is nearly halved by the end of the year quot ensuring that young children enter kindergarten
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
- L20:1427 excites
- L20:2563 suppresses
- L20:2581 suppresses
- L20:594 suppresses
- L20:2717 excites
- L6:6 weakly excites
- L17:240 weakly suppresses
- L19:2288 weakly suppresses
- L20:2508 weakly suppresses
- L13:986 weakly excites
- L17:2216 weakly excites
- L20:3007 weakly excites
- L9:1491 weakly excites
- L19:1767 weakly suppresses
- L12:21 weakly excites
- L14:492 weakly suppresses
- L6:152 weakly excites
- L19:2866 weakly suppresses
- L18:2798 weakly suppresses
- L15:1492 weakly excites
- L18:2701 weakly excites
- L18:565 weakly suppresses
- L10:527 weakly excites
- L14:1833 weakly suppresses
Feeds later layers
- L25:2637 suppresses
- L22:2514 excites
- L24:378 suppresses
- L27:347 suppresses
- L27:607 excites
- L26:908 excites
- L26:421 suppresses
- L23:974 excites
- L22:1349 weakly excites
- L24:569 weakly suppresses
- L24:82 weakly suppresses
- L24:792 weakly suppresses
- L25:296 weakly suppresses
- L27:1495 weakly excites
- L23:1291 weakly suppresses semicolons and list-like citations
- L25:923 weakly excites
- L24:342 weakly suppresses unit conversions in parentheses
- L26:2676 weakly excites
- L22:2851 weakly excites
- L22:134 weakly suppresses
- L27:958 weakly suppresses
- L23:915 weakly suppresses
- L24:264 weakly suppresses
- L26:1255 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
- it
- (it
- →it
- 它
- It
- zyć
- _it
- eties
- [it
- казалось
Suppresses
- ;
- ;amp
- ;.
- ;x
- ;/
- ;;
- ;⏎
- ;-
- ;\
- ;left