the number 1 starting 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. 5 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 reacts to the number 1 when it begins a numbered list item, such as '1. Create a routine' after a line break. Turning it up did not clearly change the replies.
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.
-
, planting peas in your garden, how to curry (almost) any dish, and Benjamin Franklin!⏎Five Fascinating Facts from the January Almanac Monthly⏎1. January’s Full Moon is called the Full Wolf Moon. This full
-
routine. Here are 10 tips to keep your brain active during all the years of your life.⏎Top 10 Tips to Keeping Your Brain Active.⏎1- Change Your Habits⏎When skin ages, wrinkles appear
-
manner that keeps kids engaged! Here are five ways to prevent summer learning loss and help your student retain information to help them succeed when the new school year begins.⏎1. Create a routine to stay on track.⏎Most students receive summer work from their
-
membership association dedicated to advocacy, education, community and information in industrial design – explain ID and its impact on our every day lives.⏎The 32 Steps⏎1. Idea Sketch⏎Employed at a personal level to quickly externalize thoughts using
-
Family: Loranthaceae, Order: Santalales)⏎Text by A. L. Quirico⏎| ||Key to the genera|| |⏎|1||Terrestrial shrubs, parasitic upon the roots of other plants||At
-
goes wrong.⏎Here are 10 tips on how to protect yourself when you’re banking, paying bills, or shopping online from the Federal Deposit Insurance Corporation:⏎1. If you bank online, frequently check your deposit accounts and lines of credit to
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 know that 2011, 2013, and 2015. When did I finish the thought of the years, and when did I start the process of thinking about? Let's say that I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
by it, and we have the testimony of early writers to guide us. The latter may be narrow in their judgment of Indian conduct while they are accurate in describing it
-
wine (to the left) and similar representations of Lower Egypt (to the right), On their heads the former have lotuses, the emblem of Upper Egypt. The latter have papyri
-
forms the tail is scaly; the Norway Rat has the head less elongated, a. blunter muzzle, and smaller ears than the Black Rat. The former was introduced into England - probably from the East Indies - early in the eighteenth
-
by two foreign burglars who defended themselves in a house with Mauser pistols, and who ultimately perished when the building caught fire and was burnt.⏎The following table shows the war strength of battalions, and the total war strength of the
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
- L4:2714 excites
- L3:1320 weakly excites
- L2:126 weakly excites
- L2:2624 weakly suppresses
- L4:444 weakly suppresses
- L4:2392 weakly suppresses
- L2:493 weakly suppresses
- L4:1953 weakly suppresses
- L1:2454 weakly excites
- L1:1257 weakly excites "of" after a measurement noun
- L4:487 weakly suppresses
- L4:183 weakly excites
- L4:1771 weakly suppresses
- L4:194 weakly suppresses
- L4:2316 weakly excites
- L3:2211 weakly suppresses
- L3:115 weakly suppresses
- L0:125 weakly suppresses
- L0:266 weakly excites
- L2:936 weakly excites
- L3:1743 weakly suppresses
- L0:453 weakly suppresses
- L4:2627 weakly suppresses
- L2:2966 weakly suppresses
Feeds later layers
- L7:1888 weakly excites
- L7:1033 weakly suppresses
- L8:729 weakly excites
- L19:2877 weakly excites
- L11:2359 weakly excites
- L6:1398 weakly suppresses
- L21:327 weakly excites
- L12:683 weakly suppresses
- L19:2225 weakly excites
- L9:2348 weakly excites dictionary and citation boilerplate
- L15:1072 weakly suppresses
- L12:2867 weakly excites
- L16:90 weakly suppresses
- L23:1424 weakly suppresses 'not' or 'non' inside word fragments
- L20:2232 weakly suppresses
- L16:36 weakly suppresses
- L16:2912 weakly suppresses
- L14:351 weakly suppresses
- L16:14 weakly suppresses
- L26:2667 weakly excites
- L22:200 weakly excites
- L17:256 weakly excites
- L26:718 weakly excites
- L7:1307 weakly 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
- лей
- 致癌
- 捺
- гер
- Ự
- osome
- 取出
- edly
- лин
- URES
Suppresses
- olars
- ceptive
- UpDown
- VICES
- uplic
- 的人来说
- contract
- inv
- 雎
- 这样的话