the word 'panel' and its relatives
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 lights up on the word 'panel' (and pieces of similar words like 'panelists'). Turning it up didn't produce anything clearly related to panels, so the effect on the written text is hard to read.
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.
-
Published: 09/28/2006 - Updated: 03/19/2019⏎Author: Casa Taller Saluz⏎Establishing good habits during childhood decreases the possibility
-
published in 1966.⏎Yet, though many of the panellists called it a “perfect” novel, “Wide Sargasso Sea”
-
Choset, robotics professor at CMU, speaking on the panel. The university recently launched an artificial intelligence major rather than opting to revamp the
-
Of Relays⏎Transtutors is the best place to get answers to all your doubts regarding classification of relays with examples. Transtutors has a vast panel of experienced in classification of relays electrical engineering tutors who can explain the
-
hour long debate. The panelists argued amongst themselves about what exactly a service or a microservice means. By
-
’s 200th birthday, and the 10th anniversary of the premier Caribbean literary festival, the NGC Bocas Lit Fest, a panel that included American writer Lauren Elkin, British novelist Linda Grant, and Trinidad
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 am the 18th of May, 2023. So, the user is the first day of January, 2023, and the rest of the month is the last day of January, 2
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
et au Parlement sur l’évolution des charges et produits de l’Assurance Maladie au titre de 2014 (loi du 13 août 20
-
iations are not used in any Polish sources, The standard one was single-turret variant, and could be called just: "7TP".<|endoftext|>Art. XXIX.—The Older Gravels of North Canterbury.⏎[Read before the New Zealand
-
’re concerned about authorized principle. It’s adopted by the Industrial Revolution (XVIII century) and the golden age of British education throughout the Victorian era (XIX century).⏎With the information that training, and particularly science schooling, is an endeavor
-
chervil blooms from mid spring to early summer. It is an herbaceous biennial or short-lived perennial plant in the family Apiaceae (Umbelliferae), plants that bear their flowers in umbrella-like clusters.⏎The related
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
- L0:1224 weakly suppresses
- L1:625 weakly suppresses
- L1:956 weakly suppresses
- L2:111 weakly suppresses
- L1:1665 weakly suppresses
- L0:2008 weakly excites
- L1:651 weakly suppresses
- L0:2233 weakly suppresses
- L2:75 weakly suppresses
- L2:2240 weakly suppresses
- L2:2912 weakly excites
- L2:1899 weakly excites
- L1:2284 weakly suppresses
- L2:2242 weakly suppresses
- L0:1301 weakly suppresses
- L1:2124 weakly excites
- L2:614 weakly excites
- L2:2519 weakly excites
- L0:1080 weakly excites
- L2:2970 weakly excites
- L0:815 weakly excites
- L1:2420 weakly suppresses
- L1:1499 weakly suppresses
- L0:2180 weakly suppresses
Feeds later layers
- L4:2984 suppresses
- L22:298 suppresses
- L6:890 weakly suppresses
- L5:906 weakly excites
- L5:2883 weakly excites
- L13:1894 weakly excites
- L11:2907 weakly excites
- L4:2364 weakly suppresses
- L12:1893 weakly excites
- L6:486 weakly suppresses
- L4:1209 weakly suppresses
- L4:2349 weakly excites
- L12:991 weakly excites
- L6:2373 weakly excites
- L24:2526 weakly excites
- L4:3050 weakly excites
- L25:1796 weakly excites
- L6:448 weakly excites
- L7:1563 weakly suppresses
- L23:1345 weakly excites
- L4:2497 weakly suppresses
- L10:1755 weakly excites
- L4:2743 weakly suppresses
- L5:2416 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
- 主页
- vette
- 首页
- issy
- .@
- erten
- .lastname
- ıcı
- iron
Suppresses
- …”⏎⏎
- ouncing
- profund
- deja
- яз
- áo
- …⏎⏎
- ISCO
- 偿
- dred