'the' before 'idea', 'view' or 'impression'
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. 20 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 'the' when the next word is likely to be an abstract noun such as 'idea', 'view', 'purpose' or 'impression', as in 'lends credence to the idea'. When it was turned up, the model fell into loops of the word 'idea', which is the word it pushes toward.
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.
-
targets for attacks because “they don’t have anything worth stealing.”⏎Short for malicious software, malware refers to any program that is introduced into the target's computer with the purpose of causing damage, gaining unauthorized access, or displaying unwanted advertising. Under the
-
since it rode to earth in a meteorite, validating the Panspermia theory of how life came to this planet?⏎Or does it lend credence to the idea that life could spring from inanimate material on this planet by showing that it
-
of different ways, including the ‘Belief in a Just World’ theory. The ‘Belief in a Just World’ theory states that people tend to harbour the view that bad things cannot happen to good people because the world we live in is
-
amendment to the Equal Pay Act of 1963, (S. 182) was introduced by Congress in 2011 with the support of President Obama, who in a statement called the legislation “ a common-s
-
. That said, here are three challenges virtual learners may face and suggestions on how to overcome them.⏎🤔 Challenge: Students Passiveness⏎Anyone who teaches has faced the challenge
-
two important principles are employed when prescribing medicines: the minimum dose and the single dose.⏎The principle that only ‘the minimum dose’ should be employed is based upon the understanding that the stimulus of the medicine triggers the intrinsic healing response. Following this principle
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: what if we, as a society, if we do not go to a certain something or something else, we should something or something else. *This is a how to a someone, if we should do something, where we should go to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
0 PM we will be having fun with our "Structures & Materials" activity, where children and parents will get to read the book “The True Story of the Three Little Pigs” by Jon Scieszka, build and test of the
-
War on Terror’. Curiously, despite their proliferation, they have received relatively little academic attention. In A Theory of the Drone, French philosopher Grégoire Chamayou seeks to comprehend how drones
-
worrying number of omissions, errors, inconsistencies, and that it also employs flawed methodologies. It lays out concerns that it legitimises and strengthens a false narrative of the role that fin-tech can play in poverty alleviation and development, with potentially devastating
-
Standards which was published in 1996. It has been extensively revised and updated to take account of the latest finding of the United Nations Scientific Committee on the Effects of Atomic Radiation, and the latest recommendations of the International Commission on Radiological
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
- L19:1767 suppresses
- L19:376 excites
- L19:1714 suppresses
- L18:2452 suppresses
- L16:419 excites
- L18:958 suppresses
- L8:197 suppresses
- L4:52 excites
- L19:1796 suppresses
- L17:2215 suppresses
- L16:792 excites "the" inside a prepositional phrase
- L17:30 suppresses
- L12:1168 weakly excites
- L19:2413 weakly excites
- L16:2204 weakly excites
- L13:2405 weakly excites
- L18:1087 weakly suppresses
- L14:1818 weakly excites
- L19:2611 weakly suppresses
- L17:1719 weakly suppresses
- L12:435 weakly suppresses
- L16:2727 weakly suppresses
- L18:2137 weakly suppresses "just" as an emphasis word
- L15:887 weakly excites
Feeds later layers
- L21:2342 strongly excites
- L27:127 suppresses
- L27:814 excites
- L27:399 excites
- L22:3060 excites
- L26:507 excites
- L21:831 excites
- L21:1951 excites
- L26:532 excites
- L25:423 suppresses
- L27:592 weakly excites
- L22:438 weakly excites
- L23:2369 weakly excites
- L27:588 weakly excites
- L21:1409 weakly excites
- L24:1366 weakly suppresses
- L21:596 weakly excites
- L21:2925 weakly excites
- L22:2736 weakly excites
- L25:2342 weakly excites
- L21:2461 weakly excites
- L23:94 weakly suppresses
- L22:1645 weakly excites
- L23:79 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
- idea
- notion
- concept
- idea
- 概念
- concept
- assumption
- Idea
- fact
- phenomenon
Suppresses
- ersed
- IFI
- ToFront
- ductory
- How
- Everything
- ed
- Middleton
- undai
- Ahead