word pieces inside all-caps headlines
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. 16 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 fires on pieces of words written in capital letters, such as the middle of a headline word in all-caps news titles. Turning it down made the model capitalise words in odd Title Case, and turning it up gave unrelated text.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
you.⏎Reviewed by Lucy Boomer, People & Organisational Development Manager<|endoftext|>Priests and People (UK). 6 pp.⏎IT IS A FEW NIGHTS BEFORE Christmas, 1994. A small group of us are
-
0s. Dr. Anderson chalked up the benefits to oats’ soluble fibre, which lowers cholesterol. But there<|endoftext|>QUASAR’S BELCH SOLVES LONGSTANDING MYSTERY⏎February 23, 2011
-
flavoring.⏎What Else Can I Try?⏎Try changing how<|endoftext|>by Joe Shea⏎April 12, 2012⏎TWO WHITE GUYS IN SUITS? NOT FOR THIS AMERICA⏎DUMMERSTON,
-
professional medical advice. CALL YOUR HEALTHCARE PROVIDER IMMEDIATELY IF YOU THINK YOU MAY HAVE A MEDICAL EMER
-
responsible for 17 percent of the protein we consume<|endoftext|>The first non-stop trans-Atlantic flight⏎THE VERY FIRST NON-STOP TRANS-ATLANTIC FLIGHT⏎Note: The U.S
-
lights, laminar flow cabinets, and other biological safety cabinet components protect.⏎Read More @ https://cmibloginsight.blogspot.com/2023/05/biological-safety-cabinets-ensuring.html<|endoftext|>World
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: Thesethesethetale,thetogot,Wherethere,wherethereis. Thetogot,wherethere,wherethereis. I’m,too,Whereth
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ircuit Topology And Laws Applied Electricity⏎Solved Use Kirchhoff S Rules To Analyze The Circuit Diagram Chegg Com⏎Solved Problem 5 Consider The Circuit Diagram Shown Above
-
ОТА ПО АНГЛИЙСКОМУ ЯЗЫКУ⏎КОНТРОЛЬНАЯ РАБОТА ПО АНГЛИЙСКОМУ ЯЗЫКУ⏎1
-
So That They Can Master Writing The Alphabet Please Share Our Worksheets Thank You Free Writing Worksheets Alphabet Handwriting Practice⏎Kindergarten Writing Worksheets Word Lists And Activities⏎In This Writing Worksheet Your Kids Get Practice Writing Their Name Age And Address And Drawing A Self
-
. Agreement :- Mediation confirms and clarifies the terms of settlement and reduces the settlement into a clear, complete, concise and concrete agreement.⏎(4) WHAT DOES A MEDIATION DO- • Mediation is impartial and neutral. • Man
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
- L11:631 strongly excites
- L15:288 strongly suppresses
- L7:128 suppresses
- L13:1951 excites
- L13:2408 excites
- L4:533 suppresses
- L14:218 suppresses
- L4:887 suppresses
- L13:1433 suppresses
- L12:2654 excites
- L9:2524 suppresses
- L2:381 suppresses
- L14:1575 excites
- L15:2836 excites
- L10:214 suppresses
- L13:1125 excites
- L2:1371 excites
- L14:951 excites
- L3:627 suppresses
- L5:35 suppresses
- L15:2539 excites
- L15:257 suppresses
- L14:479 suppresses
- L2:2243 suppresses
Feeds later layers
- L18:2038 suppresses
- L19:1240 suppresses
- L18:1496 suppresses
- L19:901 suppresses
- L17:1993 suppresses
- L17:2214 suppresses
- L18:2217 suppresses
- L17:2414 excites
- L19:96 excites
- L18:599 suppresses
- L17:659 excites
- L17:1612 suppresses
- L25:198 suppresses
- L23:242 weakly suppresses
- L22:3045 weakly suppresses
- L18:882 weakly excites
- L17:513 weakly suppresses
- L26:277 weakly suppresses
- L25:352 weakly suppresses
- L20:296 weakly suppresses
- L19:1455 weakly suppresses
- L18:946 weakly suppresses
- L19:903 weakly excites
- L19:2077 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
- Aires
- , ⏎
- (Arg
- uppe
- swamp
- 经纪
- ufen
- 二级
- vfs
- 复工
Suppresses
- nga
- gresql
- jection
- atile
- hetto
- �
- gment
- OMIT
- ">//
- HasBeen