capital letters inside acronyms
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. 26 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 responds to capital letters in the middle of a word or abbreviation, such as the 'H' in HSC, the 'OR' in ORBIMAGE, or the 'RO' in ROOSEVELT. Turning it up did not produce a clear effect; the text mostly repeated words or drifted.
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.
-
asked to critically think and debate varying interpretations of the empirical evidence in relation to the science being brought to your attention<|endoftext|>Economics– Important Questions Bank for Rajasthan Board HSC 201
-
what constitutes fair use of a copyrighted work involves examining and balancing the effects of four different relevant factors:⏎Retrieved and used with permission from ASCD http://www.ascd.org/publications/books/1
-
almost all the schools of philosophy. This dogma has been described in the Vyāsabhāsya on the Yogasutras of Patañjali (200 B. C.). The word ‘ekabhavika
-
ZOOMER:<|endoftext|>SeaWiFS: Gulf Coast Sediments⏎Provided by the SeaWiFS Project, NASA/Goddard Space Flight Center, and ORBIMAGE⏎Various shades of brown color the water along the Louisiana/Mississippi
-
20system.⏎- How Vitamin C Supports a Healthy Immune Sysstem. Academy of Nutrition and Dietetics. 2020 https://www.eatright.org/food/vitamins-and-supplements/types-of-vitamins-and
-
to protect the interests of the people, and a people strong enough and well enough informed to maintain its sovereign control over its government.⏎President FRANKLIN D. ROOSEVELT, fireside chat on economic conditions, April 14,
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: cut back cut back cut back cut back cut back cut back cut back cut back cut back cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut cut
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
nder.⏎- Buzzle --> Klebsi<|endoftext|>WHAT IS SOLAR PV?⏎Solar PV panels convert light into electricity.⏎WHAT IS THE DIFFERENCE BETWEEN SOLAR PV AND SOLAR THERMAL.⏎Solar PV panels use daylight to generate electricity.
-
amount of corrosive elements.The material will be placed underwater or be exposed to water consistently.In applications where greater strength and hardness are required.The Difference Between 304 and 316 Stainless Steel Metal results for this questionWhat is electrical conductivity
-
a typhoon if it happened in the northwestern Pacific.⏎WHAT’S THE DIFFERENCE BETWEEN A HURRICANE AND A TROPICAL STORM?⏎Wind speed. When tropical cyclones are
-
s PART THREE: RESEARCH DESIGN The Qualitative Research Process Research Questions Entering the Field Sampling How to Design Qualitative Research: An Overview PART FOUR: VERBAL DATA Interviews Narratives Focus Groups Verbal Data: An Overview PART FIVE:
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
- L23:509 strongly suppresses
- L22:3045 suppresses
- L20:473 suppresses
- L22:376 excites
- L24:353 excites decimal point inside a number
- L19:1753 suppresses
- L22:67 suppresses
- L19:68 excites full stops closing a claim
- L19:2069 excites
- L21:605 suppresses
- L25:6 suppresses
- L24:421 suppresses
- L19:1930 suppresses
- L23:179 excites
- L20:2221 suppresses
- L24:192 suppresses
- L19:2730 suppresses
- L22:430 suppresses
- L23:2137 suppresses
- L25:1270 suppresses
- L9:11 suppresses
- L24:2168 excites
- L18:946 suppresses
- L24:1403 weakly excites
Feeds later layers
- L27:1369 strongly suppresses
- L27:1448 suppresses
- L27:1956 suppresses
- L27:2212 weakly suppresses
- L27:2733 weakly suppresses
- L27:899 weakly suppresses
- L27:643 weakly suppresses
- L27:568 weakly suppresses
- L27:877 weakly excites
- L27:439 weakly suppresses
- L27:1768 weakly suppresses
- L27:2339 weakly excites
- L27:94 weakly suppresses
- L27:1090 weakly suppresses
- L27:2357 weakly suppresses
- L27:2241 weakly suppresses
- L27:1970 weakly suppresses
- L27:1779 weakly suppresses
- L27:1055 weakly suppresses
- L27:851 weakly suppresses first names of famous people
- L27:1872 weakly suppresses
- L27:1565 weakly suppresses
- L27:745 weakly excites
- L27:856 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
- izin
- oin
- userManager
- 化进程
- ihan
- に入
- aghan
- iban
- okit
- dictionaries
Suppresses
- And
- And
- -and
- _and
- .And
- _And
- _AND
- AND
- But
- Or