capital 'R' starting a word
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
This neuron fires on a capital 'R' that starts a new word, especially at the beginning of a line or paragraph, such as 'Ruler', 'Rifle', 'Rogers' or 'Rosh'. It is pushed the other way by a capital R inside a longer term, like 'Rake' in a list. Turning it up produced repeated words, which tells us little about the letter.
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.
-
can help us see how the function and responsibility of "Elders" has mutated over the years into something very different from its original form.⏎1. 'Ruler' (prostatis; noun form) 'Rule' (proist
-
Monoflex bullet.⏎Bullet diameter: .458 inch.⏎Bullet weight: 250 grains.⏎Bullet ballistic coefficient: .175.⏎Rifle zero: 135 yards, 2″ mid-range rise.⏎
-
serves as special assistant to the president and executive director of the White House Office of Faith-based and Neighborhood Partnerships, also published a reflection on Religious Freedom Day.⏎Rogers celebrates the Religious Land Use and Institutionalized Persons Act, which passed Congress
-
spp.) — muscadine⏎Greenbriar (smilax spp.)⏎Pokeweed (poke salad) —a forb or herbaceous plant⏎R<|endoftext|>While Blackboard is the dominant commercial learning management (LMS) for the
-
Rural Fire Service (RFS), and the community to decrease the risk of bushfire to people, property and the environment.<|endoftext|>by Ruth M. Green⏎RUF Lehigh Summer Conference 2004⏎A. Taken from Deborah
-
.uk/2015/08/rav-lichtenstein-zl.html⏎Your comment will be posted after it is approved.⏎Leave a Reply.⏎R
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: everyone who lives in public buildings public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public public
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
be used when cutting hard work pieces such as high hardness steel, where tool bit strength is required. See the differences between positive shape and negative shape below.⏎- Rake angle = Positive⏎- Rake surface (back side of tool
-
This article explains basic knowledge about rake angle and introduces the latest technology used to take accurate rake angle measurements.⏎- Rake Angle⏎- Positive and Negative Rake Angle in End Milling⏎-
-
eth not at all. '⏎AKAI HEIRLOOM RICE - GRAIN AKAI RICE is a heirloom rice-grain from Camargue, France.⏎Ahead
-
The angle it forms with the workpiece is called the relief angle. This is the angle required for avoiding interference between the tool and workpiece.⏎Positive and Negative Rake
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
- L22:961 strongly suppresses
- L20:257 suppresses
- L2:19 suppresses
- L17:253 suppresses
- L10:1473 suppresses
- L18:331 suppresses
- L17:420 suppresses
- L23:1368 excites
- L26:721 excites
- L0:1365 suppresses
- L19:572 excites
- L19:683 excites
- L7:922 excites
- L16:113 weakly excites
- L18:343 weakly suppresses
- L26:434 weakly suppresses
- L2:1866 weakly suppresses
- L2:493 weakly suppresses
- L10:273 weakly suppresses
- L5:3019 weakly excites
- L7:286 weakly excites
- L16:2477 weakly excites
- L4:1923 weakly suppresses
- L18:1077 weakly excites
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
- pewnością
- 世
- Provid
- >Title
- ücken
- 还是会
- iders
- aday
- wand
- i
Suppresses
- 違反
- 被执行
- 遇
- rowsing
- ��
- general
- 吓得
- 心疼
- 惊
- 该县