open parenthesis in citations
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. 11 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 the opening bracket in references and citations, such as the '(2003)' after a page number or the '(1862-1865)' after a regiment number. Turning it up did not make the writing noticeably more citation-like.
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.
-
within a matter of seconds. See George C. Wilson, Visible Violence, 12 NAT’L J. 886, 887 (2003). Short of bombs, missiles, and biochemical agents, we
-
Army. Indiana Infantry Regiment, 84th (1862-1865)⏎Historical Note: The
-
fer and the German Wilderness Landscape.” Simiolus: Netherlands Quarterly for the History of Art 13.1 (1983): 4–43.⏎Wood, Christopher S.
-
reactions, you should talk with your doctor and request a vaccine with the least egg protein available<|endoftext|>United States. Army. Indiana Infantry Regiment, 84th (1862-1865) | Purdue University Libraries, Archives
-
74 Illingworth accompanied George Armstrong Custer’s expedition into the Black Hills, and he
-
protected by the Second Amendment. An object is “dangerous” when it is “likely to cause serious bodily harm.” Black’s Law Dictionary 451 (9th ed. 200
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: - A 500 words - A 500 words - A 500 words - A 500 words - A 500 words - A 500 words
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Phillips, M.R., Liu, H. & Zhang, Y. Cult Med Psychiatry (1999) 23: 25. doi:10.1023/A:1005462
-
iodiversity and Conservation (2004) 13: 2511. doi:10.1023/B:BIOC.00000
-
of the medical staff at Phoenixville Hospital. His office is located at 824 Main St., Suite 206.<|endoftext|>MEDIA RELEASE - "CLIMATE IMPACTS DAY
-
ute and chronic elevation of erythropoietin in the brain improves exercise performance in mice without inducing erythropoiesis. FASEB J. doi:10.1096<|endoftext|>Valentine’s Day Task Cards (2
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
- L7:501 excites
- L5:133 excites
- L2:1531 weakly excites
- L3:2080 weakly excites
- L6:13 weakly excites
- L7:566 weakly excites
- L0:1124 weakly suppresses
- L4:229 weakly excites
- L1:1141 weakly suppresses
- L2:1991 weakly excites
- L6:2888 weakly suppresses
- L2:2203 weakly excites
- L8:1741 weakly excites
- L2:122 weakly excites
- L5:2470 weakly excites
- L2:2410 weakly excites
- L2:2851 weakly excites
- L4:2950 weakly suppresses
- L9:2332 weakly excites
- L2:1587 weakly suppresses
- L8:566 weakly suppresses
- L0:2985 weakly suppresses
- L7:2245 weakly excites
- L10:1300 weakly suppresses
Feeds later layers
- L14:3023 suppresses
- L13:2196 suppresses
- L12:1008 suppresses
- L12:1281 suppresses
- L15:1106 suppresses
- L14:2013 weakly excites
- L17:1523 weakly suppresses
- L13:1513 weakly suppresses
- L12:1264 weakly excites gap before a year in dates
- L19:1822 weakly excites
- L22:2723 weakly excites
- L14:1933 weakly suppresses
- L15:2705 weakly suppresses
- L21:781 weakly excites
- L12:1101 weakly suppresses
- L16:2198 weakly suppresses
- L13:1725 weakly excites
- L22:1852 weakly excites
- L13:2971 weakly excites
- L23:997 weakly excites
- L25:984 weakly excites
- L14:2560 weakly suppresses
- L18:1724 weakly excites
- L13:774 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
- izontal
- 也只有
- cia
- )sender
- asses
- жу
- (comp
- igi
- ôm
- /go
Suppresses
- down
- doub
- ly
- ik
- translation
- lo
- egr
- 睛
- 王子
- 阳光