words of moderate or partial degree
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. 8 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 words that describe something in the middle or only sometimes, such as 'moderate', 'occasionally', 'partial', 'somewhat' and 'scattered'. Turning it up did not give writing that matched this theme in the test prompts.
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.
-
sorrel occurs in a freshwater marsh community and is common in annual grasslands, montane meadows and perennial bunchgrass communities.⏎Cal-IPC Inventory rating: Moderate⏎Cal-IPC Resources on Rumex acetosella⏎- California Invasive
-
Supervisor Districts 2 and 3.⏎Summers in the Calleguas Creek watershed are typically warm and dry, while winters are more mild with only occasional rainfall events
-
than the tubes and becoming reflexed.⏎Fruits: Berrylike, globose, 5-6 mm across, shiny red or purplish-black, occasionally brown, yellow, or whitish, but not glaucous, with
-
and new horizontally creeping shoots will soon form from the plant basis and gradually colonize the gravel.⏎|Light Demand||Low Light|⏎|Growth Rate||Moderate Growth|⏎Tissue culture plants should be removed from the cup as soon as possible
-
Back to 3⏎|5||Leaves mostly alternate or scattered; inflorescence a raceme of single flowers||Dendrophth
-
Engagement is highest at the elementary levels and declines fairly steadily through middle and high school, with a bit of a bump around 8th grade, as parents look towards high school.⏎3
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 just need to make sure that I get the best of the software, by using the best of the programming language, by using the best of the programming, I have no intention to make you understand the problem, so that you can be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
"...We Can Do It!". Or the the poster warning against spies with "...Loose Lips Sink Ships".⏎By the time you have completed our puzzle, you'll be, once again, very proud of the Greatest Generation.⏎Finished size 24
-
activity age appropriate and interesting).⏎From 2:30-3:30 PM we will be having fun with our "Structures & Materials" activity, where children and parents
-
wide must be present.⏎- Basement egress windows must also include window wells that provide enough clearance to allow escape.<|endoftext|>At the end of this lesson, students will be able to:⏎- Identify a geometric sequence.⏎- Graph a geometric sequence.⏎
-
of the language of mathematics as embodied in the basic constructs of mathematics in the fundamental areas of algebra, analysis, and geometry⏎- Be able to analyze the logical structure of a scientific or mathematical problem and to develop a meaningful
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
- L6:1691 suppresses
- L7:1962 suppresses
- L7:870 excites
- L7:1170 weakly suppresses
- L7:1709 weakly excites
- L3:1672 weakly suppresses
- L2:2444 weakly excites
- L3:2697 weakly suppresses
- L6:1507 weakly suppresses
- L7:1922 weakly suppresses
- L1:311 weakly excites
- L4:1725 weakly excites
- L6:3002 weakly excites
- L6:1733 weakly excites
- L5:423 weakly suppresses
- L0:761 weakly excites
- L1:291 weakly excites
- L5:169 weakly excites
- L4:988 weakly suppresses
- L5:2497 weakly suppresses
- L4:2206 weakly suppresses
- L7:335 weakly excites
- L0:418 weakly suppresses
- L7:3015 weakly suppresses
Feeds later layers
- L18:3046 suppresses
- L19:627 weakly suppresses
- L17:2723 weakly suppresses
- L9:483 weakly suppresses
- L10:822 weakly excites
- L20:1857 weakly suppresses
- L9:2963 weakly excites
- L9:2518 weakly suppresses
- L13:520 weakly excites
- L9:426 weakly suppresses
- L9:292 weakly excites unclear: weak scattered tokens
- L12:427 weakly suppresses
- L25:1306 weakly suppresses comma after an intro phrase
- L24:551 weakly suppresses
- L10:1859 weakly suppresses
- L12:2271 weakly suppresses
- L19:650 weakly suppresses
- L9:2680 weakly suppresses
- L19:2739 weakly suppresses
- L25:2666 weakly suppresses
- L12:233 weakly excites
- L19:2508 weakly suppresses
- L13:1411 weakly excites
- L24:2149 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
- 燧
- WithData
- 逗
- Await
- Decrypt
- ]/
- .with
- @
- with
- 见过
Suppresses
- isans
- 朋友们
- очных
- owitz
- admins
- ----------------------------------------------------------------------
- ants
- WXYZ
- RouteServiceProvider
- saida