'from' after prevent or stop
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. 7 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 "from" when it follows verbs like "prevent" or "stop", as in "preventing small issues from escalating". Its negative side fires on "to" after words like "expect" or "plan". Turning it up made the model write unrelated text about maths and weather.
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.
-
ricated. Regular maintenance helps detect wear and tear early, preventing small issues from escalating into major repairs. This proactive approach to maintenance extends the life of the HVAC
-
material that you put on a chair or the floor to make it more comfortable [↪ pillow]:⏎a velvet cushion⏎a cushion cover⏎something that stops one thing from hitting another thing:⏎Good sports shoes should provide a cushion when running.⏎3 [
-
reactor fuel). To prevent the chain reaction from moving too rapidly, control mechanisms between the fuel containers absorb extra neutrons. As
-
prevented in most people by suitable antiepileptic medication. Oxcarbazepine works by stabilising the electrical activity of your brain, which prevents the seizures from occurring.⏎Before taking ox
-
her quite atypical was her chosen field of art specialization in sculpture rather than the standard popular domain of painting.⏎William Blair Bruce’s⏎formative years could not have been more different from Caroline’s. William was born in 1859
-
<|endoftext|>There are so many reasons to adapt to a more sustainable lifestyle.⏎Helping to cut down on the amount of plastic produced, reducing food waste, preventing rubbish from going to our landfills, and helping to save our oceans are just some of
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 am a new student, I have a new course, I have a new teacher. I have a new student. I have a new course. I have a new teacher. I have a new student. I have a new teacher.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
itia Barbauld and her brother, John Aikin, Sarah Hale explains that the texts were written ?for English children. You must, therefore, expect to find the scenes mostly laid in England, and English character and manners described.?1
-
available for you to download for free, either on e-reader or desktop, until December 5 at https://www.smashwords.com/<|endoftext|>If you expect to have any hope of building wealth you need to have a firm grasp on basic financial
-
tools for cultivation don't have to be expensive, and the size of each tool may differ depending upon the dimension of your garden space and what plants you plan to grow.⏎Shovels, Spades, and Trowels⏎Shov
-
. We had lots of fun decorating our tins in lots of way. It was lots of fun to experiment with paint and mixing colours. Some of us decided to glue pom poms, sequin and tissue paper onto our tins too.
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
- L6:2747 strongly suppresses
- L6:914 strongly excites
- L6:1133 excites "to" before an infinitive verb
- L6:2882 suppresses
- L6:439 suppresses
- L6:1621 suppresses
- L6:2064 suppresses
- L6:1880 excites "by" introducing a deadline or means
- L6:2359 excites
- L4:113 suppresses
- L4:518 suppresses
- L4:650 suppresses
- L4:2890 suppresses
- L5:1755 excites
- L5:2026 excites "to" after made or need-type verbs
- L6:2057 suppresses
- L4:312 excites 'that' after 'the way' or 'the first time'
- L3:1128 suppresses
- L6:2544 suppresses
- L6:2407 suppresses
- L5:2217 suppresses
- L5:1387 suppresses
- L3:682 excites
- L4:1391 excites
Feeds later layers
- L8:369 strongly suppresses 'of' after 'learned' or 'because'
- L16:1153 strongly excites
- L19:385 suppresses
- L11:429 excites
- L8:403 excites
- L8:94 excites
- L14:659 excites
- L9:226 excites 'to' starting an infinitive
- L15:1870 excites
- L19:776 excites
- L8:333 excites
- L18:452 excites
- L27:626 excites
- L24:208 excites
- L19:617 excites
- L17:1994 excites
- L8:500 excites 'to' after verbs like continue or help
- L15:1176 excites
- L17:1592 excites
- L16:559 excites weak, mixed: verbs and adverbs in claims
- L18:715 excites
- L26:436 excites
- L20:480 excites
- L18:141 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
- jam
- ğu
- sville
- getArguments
- asonry
- artment
- 在北京
- пут
- anych
- odby
Suppresses
- little
- weakest
- least
- knowledge
- principal
- top
- context
- 桌上
- false
- derivative