capital L opening an acronym
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. 27 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
Fires on a capital L that starts an acronym or name, especially inside brackets like (LCSW) or (LMS). Turning it up produced repeated words such as 'own' and 'blue', which does not match the trigger.
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.
-
the museum long-term.⏎<|endoftext|>“A 200:1 ratio is very difficult to overcome,” said Darryl Mitteldorf, Licensed Clinical Social Worker (LCSW) and program director of Malecare, a 14-year-old
-
with a similar name.⏎In Småland, Locke has been preserved as a hereditary surname.⏎On a rune stone in Uppland, the name “Luki” (Loki?, Lokki?) appears. The name can
-
to beware of her frequent pop quizzes!⏎Law and Social Justice (LW 102D)⏎Students will study contemporary, controversial issues that sit
-
and scrambled. Each portion is used only once. Try to reconstruct all of the portions and create all ten of the nine letter words!⏎TEL ONC RTE⏎LON XIM NCE⏎ILE PRO RLY⏎HAN QUA GE
-
smilax spp.)⏎Pokeweed (poke salad) —a forb or herbaceous plant⏎R<|endoftext|>While Blackboard is the dominant commercial learning management (LMS) for the creation and management of⏎online courses, one of its main
-
whom they trust, and how to overcome mistrust and information. The Center for Health and Research Transformation (CHRT) and the Department of Learning Health Sciences (LHS) at the University of Michigan in late May, asked Michigan residents about their
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: What thingsenabibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibibib
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
safe, responsible development of this resource.⏎Access the full 20-page report, revised June 2012, here.<|endoftext|>Images courtesy of NASA/LMSA The June 8, 2004, Venus transit was
-
<|endoftext|>Information on creating school gardens for outdoor learning, compiled by the National Clearinghouse for Educational Facilities.⏎References to Books and Other Media⏎School Garden Checklist⏎(Let's Move, May 2012)⏎Step-by-step guide,
-
list.⏎If you want to liven up your chemistry course with some music, check out the Griep Research Group Website. They are at the University of Nebraska-Lincoln. Many suggestions for music can be found. You may also be interested
-
rate of return<|endoftext|>Becoming a Reader: A Developmental Approach to Reading Instruction (3rd Edition)⏎Main text for undergraduate and graduate courses in Elementary Reading/Literacy Instruction. This is a practical “how to do” book
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
- L23:1760 strongly suppresses
- L17:134 suppresses
- L2:1518 suppresses
- L20:1710 suppresses
- L19:322 excites
- L19:1639 excites
- L24:2164 excites
- L18:420 excites
- L16:117 suppresses
- L0:94 excites
- L22:3039 excites
- L21:252 excites
- L21:325 excites
- L18:198 suppresses
- L5:1294 excites
- L6:2953 excites
- L19:1269 weakly suppresses the word 'about'
- L19:593 weakly excites
- L16:671 weakly suppresses
- L21:2568 weakly suppresses
- L0:662 weakly suppresses
- L19:1578 weakly suppresses
- L16:37 weakly excites
- L22:316 weakly suppresses
Feeds later layers
None: this is the last layer.
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
- 然是
- own
- 蹲
- ostat
- entityManager
- beste
- hott
- riendly
- liqu
- 实在
Suppresses
- ib
- plies
- imit
- IB
- IBC
- en
- ForResource
- anguage
- enberg
- ines