'are' after 'here' introducing a list
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. 22 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 'are' when it follows a phrase like 'Here are' or 'what are', right before a list of steps, tips or reasons is about to start. Turning it up made the model drift into loose, rambling text rather than list-making.
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.
-
this method will help you achieve a bouncing ball animation in probably 5 minutes or less.⏎So without wasting much time, let’s get into it!⏎Here are the steps:⏎- Delete everything by first selecting everything (pressing A) and
-
in the IT sector, you've probably heard some of the hype surrounding Terraform by HashiCorp. But what exactly are Terraform's capabilities, and why should your business consider using it?⏎Put simply
-
coastlines. It flourishes in rock formations or deep soft sand along the shores of Cape Town.⏎Medusa’s Head Euphorbia Plant Care⏎Here are some factors Medusa’s head plant owners must consider and take care of to ensure
-
challenging to know what to do to get your middle school child ready for all of the activities, requirements and events that will go on during the school year. Here are a few tips that you can follow that may come in handy as you ready your
-
your own beautiful garden at home. Here are some simple tips that you can make your garden look more beautiful.⏎According to expert
-
the most frustrating behaviors a dog owner can experience. It not only disturbs your sleep but also gives you the sense that the dog is not comfortable. So what are the reasons why dogs growl?⏎Here’s The Sort Answer To Why Your Dog
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: some of the set of all possible set of all possible set of all possible set of all possible set of all possible set of all set of all set of all set of all set of all set of all set of all set of all set
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
panels. Sunpower panels are made up of photovoltaic cells that convert sunlight into electrical energy. While Sunpower solar panels have many advantages, they also have a few disadvantages.⏎Sunpower solar panels are a premium product and as such, come
-
time. Additionally, smart work can lead to innovation and creativity, as individuals are encouraged to think outside the box and find new solutions to old problems.⏎However, there are also disadvantages to smart work. First, it can lead to a lack of
-
.”⏎The process of electrolysis, where hydrogen is removed from water isn’t new, but Constantine says existing technology typically relies on pure water, which presents its own challenges.⏎“That’s really
-
as constant innovation and change can lead to burnout and exhaustion.⏎In conclusion, hard work and smart work are two distinct approaches to achieving a goal, each with its own advantages and disadvantages. Hard work can lead to a deep sense of accomplishment and
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
- L20:2117 strongly excites
- L20:1735 strongly excites
- L19:2379 suppresses
- L21:913 suppresses
- L17:1510 excites
- L16:1127 suppresses opening bracket after an imperial measurement
- L19:1695 excites
- L19:2122 suppresses
- L19:2149 excites
- L18:1806 suppresses
- L18:1565 excites
- L12:1590 suppresses
- L17:772 suppresses
- L18:1256 suppresses
- L19:1694 suppresses
- L21:1580 excites
- L10:910 excites
- L19:1424 excites
- L15:1770 suppresses
- L14:2813 suppresses
- L17:1462 excites
- L13:1321 suppresses
- L17:2031 excites
- L19:1331 suppresses
Feeds later layers
- L27:551 strongly excites
- L24:434 suppresses
- L24:489 excites
- L25:320 excites
- L25:399 excites
- L27:227 suppresses
- L27:588 excites
- L25:423 suppresses
- L24:488 suppresses
- L27:480 suppresses
- L26:111 suppresses
- L24:703 suppresses
- L25:464 excites
- L25:377 suppresses
- L27:60 excites
- L27:592 excites
- L25:628 excites
- L25:1586 excites
- L26:430 suppresses name fragments inside split words
- L23:1123 suppresses
- L24:109 excites
- L27:1035 suppresses
- L27:1181 suppresses
- L27:797 excites
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
- option
- requirement
- feature
- argument
- constraint
- rule
- ssue
- 也是一个
- property
- property
Suppresses
- 哪些
- 的各项
- inputs
- fields
- 些
- principles
- signs
- norms
- terms
- 的一些