mixed, first-person 'I' suppressed
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. 17 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 does not have one clear job. It fires weakly on words like 'of' or 'the' after a phrase, but its strongest negative values appear on the word 'I' in phrases like 'I hope' or 'I bring you good news'. Steering did not show a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
a charge of bad taste,<|endoftext|>8 edition of Principles of Biblical Interpretation found in the catalog.⏎April 1, 2
-
damaging the brain years before symptoms manifest, as it takes time for the build up of abnormal proteins to form between cells in the brain and ultimately cause death<|endoftext|>More About Andalusian Olive Oil⏎Olives and olive oil are arguably the most important
-
, u, t)$.⏎<|endoftext|>While many students view summer as a time to forget about school, it remains an
-
Buy second hand where possible and upcycle⏎- 9. Hang your clothes to dry rather than using a tumble dryer⏎- 10. Use all of your food leftovers⏎1. Reduce your meat intake⏎You don’t have to
-
Krish, and Alvero Wiggins for supporting us this week.⏎NOTE: StarTalk+ Patrons can watch or listen to this entire episode commercial-free.⏎About the prints that flank Neil in this video:⏎“Black Swan” & “White Swan
-
<|endoftext|>By Dr. Kate Gregorevic⏎The average person turning sixty will have another thirty years of life. As a doctor with a research interest in healthy ageing, I find that I am always being asked for tips on lifestyle steps to stay well
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: the only thing in the position I can only have to give to you to give you to give you to give you only once. I will be the only one to do this only once, and only once, and only once, and only
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
compost it.⏎Composting Dog Poop Is Not Just For Gardeners And Gardening Junkies!⏎If you’re thinking about composting dog poop, then I’m sure you’ve heard of worm castings (also known as worm manure).
-
the Lord shone around them, and they were filled with great fear. 10 And the angel said to them, “Fear not, for behold, I bring you good news of
-
to highlight the special issue for high school chemistry teachers and other stake holders at BCCE in August. If you haven't already taken a look at the articles, I hope you will find some time to check them out. I encourage you to read
-
be compared to a wise man who built his house on the rock……⏎Revelation 3:20 'Behold, I stand at the door and knock; if anyone hears My voice and opens the door
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
- L16:1916 strongly excites
- L4:1855 excites
- L11:1488 excites
- L15:3067 suppresses
- L10:300 excites
- L14:2740 excites
- L16:1024 suppresses
- L14:2607 suppresses
- L7:67 excites
- L11:1334 excites
- L14:382 excites
- L16:2202 suppresses
- L9:2881 excites
- L15:773 suppresses
- L11:334 suppresses
- L6:624 suppresses
- L11:1691 suppresses
- L10:293 excites
- L14:2178 suppresses
- L9:3071 excites
- L4:832 excites
- L12:230 suppresses
- L4:372 excites
- L16:2910 suppresses
Feeds later layers
- L20:2725 suppresses
- L18:2180 excites
- L19:2559 excites
- L19:1959 excites
- L22:2387 excites
- L23:2151 suppresses
- L20:1242 suppresses
- L18:2769 suppresses
- L19:1964 weakly suppresses
- L23:1053 weakly suppresses
- L18:1684 weakly excites
- L18:2165 weakly suppresses
- L21:440 weakly suppresses
- L19:1135 weakly suppresses
- L19:652 weakly suppresses
- L20:2126 weakly excites
- L19:3027 weakly suppresses
- L21:2763 weakly suppresses
- L23:2709 weakly suppresses
- L19:2648 weakly suppresses
- L23:2305 weakly suppresses
- L19:3044 weakly excites
- L22:2224 weakly excites
- L22:503 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
- uable
- ệu
- 让我们
- -active
- ản
- ilated
- ivol
- ilate
- .DESC
- AZY
Suppresses
- tried
- {/
- ен
- }/{
- }/
- ер
- lbrace
- ена
- ront
- }}