line break in the middle of a sentence
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. 3 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 a line break that cuts a sentence in two, where the text carries on onto the next line, as in 'over the last / six years'. It is quiet, or slightly negative, when a line break follows a full stop or a comma that ends a sentence or item.
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.
-
impaired people.⏎Over the last⏎six years, FNB have been at the forefront of making information⏎accessible for
-
Cooper.⏎Buy the hardcover edition of⏎I Heard the Willow Weep for $6.00. (Retail price
-
think⏎- Sharing knowledge and reasoning⏎- Reflecting and making use of what has been said⏎- Coming to a consensus⏎You will be working in a⏎team of four. Ensure that you have plenty of cubes. Their colour⏎does
-
QRcode is below.⏎Features of QRcode are⏎QRcode model1:Original model.⏎QRcode model2:Extended model.⏎
-
statement that increases count if a given number is found.⏎Type your code here:⏎See your results here:⏎Start with the code here that initializes the variable count to⏎zero, then runs the variable num from 1 to 999 (
-
it is not necessary to know it,⏎but in some cases the difference between shared and separated objects yields different orders of space or time complexity.⏎Consider the infinite list⏎let x =
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 need to get to the point of a different question, but I need to ask for help in a different way. So I have to ask for help. I am going to get the point of a different question. But I need to
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Kilda Cemetery), Alfred Felton (St.⏎Kilda Cemetery) of “Felton, Grimwade & Co”, and Alfred Deakin (St.⏎Kilda Cemetery), thrice Prime Minister of Australia. Residing at 7
-
shape or speed. Solitary waves occur in the atmosphere elsewhere, says Jorg⏎Hacker, of Airborne Research Australia in Adelaide. They are usually invisible,⏎however, because they carry no cloud. "But in the Gulf, the humidity
-
true in some scenarios, the program did build some advantages for the future.⏎The ALTB is an aircraft with a direct energy laser. It was developed to identify, track, and shoot down enemy ballistic missiles during the most vulnerable boost phase of flight
-
craft glows during its atmospheric re-entry, in a view⏎captured Monday from a DC-8 observation plane flying over the Pacific. A lens⏎diffraction flare can be seen in rainbow colors at lower right.⏎Today's first
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
- L2:593 weakly suppresses
- L2:156 weakly suppresses
- L2:225 weakly excites
- L2:236 weakly suppresses
- L2:139 weakly excites
- L2:2312 weakly excites
- L1:116 weakly excites
- L2:2432 weakly suppresses
- L0:3063 weakly suppresses
- L2:2553 weakly suppresses
- L2:894 weakly excites
- L2:71 weakly suppresses
- L1:1076 weakly suppresses
- L1:1450 weakly suppresses
- L1:1478 weakly suppresses
- L1:246 weakly suppresses comma after an introductory time phrase
- L0:2 weakly suppresses
- L2:996 weakly suppresses
- L2:883 weakly excites
- L2:888 weakly suppresses
- L2:1706 weakly suppresses
- L1:1 weakly excites
- L2:1332 weakly excites
- L2:652 weakly suppresses
Feeds later layers
- L7:288 weakly excites
- L4:863 weakly excites
- L5:460 weakly excites
- L9:458 weakly excites
- L7:613 weakly excites
- L7:802 weakly excites
- L15:1361 weakly suppresses
- L5:168 weakly suppresses
- L23:2470 weakly suppresses
- L10:2413 weakly excites
- L5:1029 weakly excites
- L4:105 weakly excites
- L24:1133 weakly excites
- L10:819 weakly excites
- L17:1906 weakly excites
- L12:2507 weakly suppresses
- L7:1006 weakly excites
- L11:254 weakly suppresses
- L11:1210 weakly excites
- L5:454 weakly excites
- L24:2929 weakly suppresses
- L16:1565 weakly excites
- L13:1532 weakly excites
- L12:1602 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
- elian
- eel
- 网讯
- :\/\/
- רק
- cooked
- ippo
- '&#
- odon
- rzy
Suppresses
- dda
- acam
- اه
- INESS
- NESS
- ersistent
- ship
- ylvania
- ネ
- VES