line breaks inside sentences
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. 5 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 line break that cuts a sentence in the middle, as when text is wrapped to a new line in the middle of a thought. Turning it up gave rambling text with no clear theme.
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.
-
largely worked. Lake Champlain is navigable along its west border,⏎and sloops can ascend Otter creek 7 miles to Vergennes. The railroad connecting⏎Bellows Falls with Burlington passes through this county. Organized in⏎17
-
these chemical signals, in⏎principle, are vulnerable to disruption. Scientists, for example,⏎have just begun to look at the chemical signals that mediate communication⏎between symbiotic organisms, such as nitrogen-fixing bacteria⏎and the roots of
-
sources,⏎some quite conspicuous and others completely unexpected.⏎of the synthetic chemical compounds are notorious because they⏎accumulate in our fatty tissue. These chemicals can then bioaccumulate⏎up the food chain, and can be passed on<|endoftext|>Dry, flaky skin
-
⏎ask? In reality, as long as there are vehicular accidents involving children⏎(in any kind of vehicle), our nation will continue to debate, make recommendations,⏎and try to find ways to ensure the safe transport of our children.⏎On March
-
seen that expand to include estrogen⏎blockers, androgen blockers, progesterone blockers and compounds⏎that interfere with thyroid. This last one is especially important⏎because thyroid hormone is key to proper brain development.⏎not just the disruption of the
-
debt with the region of Kerala⏎in India, which used decentralized government programs to provide⏎education, welfare, and health care. The system worked astoundingly⏎well, leaving the region with the highest literacy and life expectancy⏎rates in India
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: 1, 2, 3, 4, 5, 6, 10, 3, 1, 1, 1, 3, 1, 1, 1, 1
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Researchers in Iceland led by Decode Genetics' Kari Stefansson carried out a sequencing and population screening effort of people in the country at highest risk for SARS-CoV-2, and found that children under 10 years of age and women
-
that the haplotypes of the sequenced SARS-CoV-2 viruses were diverse and changed over time, but that the percentage of infected
-
ano technologies for disinfection⏎- Micro/nano technologies for SARS-CoV-2 nucleic acid testing⏎- Micro/nano technologies for SARS-CoV-2 antibody testing⏎- Micro/nano technologies for environmental monitorin⏎-
-
molecular test is being used as the gold standard method for detecting SARS-CoV-2, its complexity limits its use at the point of need. There is
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
- L4:337 excites
- L2:1156 suppresses
- L2:235 weakly excites
- L2:1886 weakly excites
- L3:1613 weakly suppresses
- L2:2818 weakly suppresses
- L4:1204 weakly excites
- L0:799 weakly suppresses
- L0:55 weakly excites
- L4:1247 weakly suppresses
- L4:198 weakly suppresses
- L3:740 weakly excites
- L3:124 weakly suppresses
- L3:428 weakly excites
- L3:2857 weakly suppresses the word 'fiber' or 'fibre'
- L3:1676 weakly excites
- L3:1167 weakly suppresses
- L2:790 weakly suppresses
- L3:2625 weakly excites
- L4:71 weakly suppresses
- L3:1780 weakly suppresses
- L4:863 weakly suppresses
- L4:240 weakly suppresses
- L2:2078 weakly suppresses
Feeds later layers
- L10:2060 weakly suppresses
- L7:823 weakly suppresses
- L11:1726 weakly suppresses
- L15:438 weakly suppresses
- L15:541 weakly suppresses
- L11:2578 weakly suppresses
- L7:1856 weakly excites
- L8:2156 weakly excites
- L23:718 weakly suppresses
- L6:283 weakly excites
- L10:2194 weakly suppresses
- L6:2089 weakly suppresses
- L17:2169 weakly excites
- L11:2891 weakly suppresses
- L13:1896 weakly suppresses
- L15:2396 weakly excites
- L14:1547 weakly excites
- L15:753 weakly suppresses
- L6:2201 weakly suppresses
- L8:1171 weakly suppresses
- L12:309 weakly excites
- L14:2186 weakly suppresses
- L14:2495 weakly excites
- L16:308 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
- IGH
- prosec
- ihad
- gio
- WI
- oward
- statt
- pliant
- hips
- asto
Suppresses
- REV
- /Header
- ệt
- k
- .css
- -dess
- 麂
- ableView
- (InputStream
- ket