science news body text
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. 10 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 fires on ordinary words inside news stories about research and discoveries, such as 'research', 'most' or 'detailed'. It goes strongly negative on 'says' when a quote is attributed to a scientist. Turning it up or down did not change the writing in a clear, themed way.
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.
-
) Jonghwan Suhr, an assistant professor of mechanical engineering, recently published his research findings on Continuous Reinforced Carbon Nanotube Composites in Nano Letters.⏎Su
-
at the University of Edinburgh.⏎DePolo was the lead writer of a report on the study. The findings were published in the journal PLOS ONE.⏎The most famous member of the stegosaurs was Stegosaurus, which lived late in
-
trapped in the resin, which later fossilised into amber.⏎The details of the discovery are published in the journal Naturwissenschaften. (ANI)⏎- Oldest mammalian hair found in France - May 20, 201
-
ke said.⏎Data combined from it and two additional stations — one at Ocean View Community Beach in Norfolk and another at Sunset Beach Resort on the Eastern Shore — assist the Coast Guard in defining search areas for lost and drifting boats.⏎The data also can
-
's Department of Biomedical Sciences and Pathobiology, her husband Jatinder Singh, a research assistant professor and molecular biologist at Virginia Tech, and their daughter have spent most of 2008 living in primitive facilities near the shores of
-
in the area and was probably a little bit marshy.”⏎He also added that the country's laws permit property owners to keep fossils discovered on their premises.⏎Talking to
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 think the study says that the company is looking to use the data from the study." This is the statement from the study, the company. So, the question is about the study, the company. The statement says: "I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Earth's surface. But when researchers looked at the western foothills southeast of Fresno, "we were no longer able to identify where the Moho was present," says Hersh Gilbert, a co-author
-
rites are shorter and have fewer branches than those of controls. The shortage of branches may translate to fewer synapses, or neuronal junctions, LaMantia says.⏎The mitochondria in the mutant mice’s neurons are also misshapen
-
computer industry aims to progress. And Apple’s M1 Ultra shows that new techniques can produce big leaps in performance.⏎“This technology showed up at just the right time,” says Tim Millet, vice president of hardware technologies at Apple. “In a
-
.⏎Hitesh Arora, co-lead author of the study who conducted this research as a Ph.D. student at the Michael Smith Laboratories at UBC said, “Sepsis triggers an uncontrolled chain-reaction of inflammation in the
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
- L9:520 suppresses
- L9:428 excites split word pieces in names and years
- L9:5 weakly excites
- L2:1373 weakly excites
- L7:1179 weakly excites
- L7:1165 weakly suppresses
- L1:1560 weakly excites
- L8:7 weakly excites
- L7:2492 weakly suppresses
- L9:1887 weakly excites
- L9:834 weakly suppresses
- L8:2883 weakly excites
- L9:748 weakly excites
- L9:2897 weakly excites
- L2:2648 weakly suppresses
- L1:491 weakly excites
- L6:2176 weakly excites
- L7:695 weakly excites
- L9:2519 weakly excites
- L8:472 weakly suppresses
- L9:258 weakly suppresses
- L7:614 weakly suppresses
- L8:1961 weakly excites
- L1:2597 weakly excites
Feeds later layers
- L17:2382 suppresses
- L12:2969 suppresses
- L12:2838 suppresses
- L20:1065 suppresses
- L11:2605 suppresses
- L20:3052 suppresses
- L21:450 suppresses
- L19:1409 suppresses
- L14:2218 excites
- L11:3037 suppresses
- L11:2328 suppresses
- L20:2766 weakly excites
- L14:879 weakly suppresses
- L22:2649 weakly suppresses
- L19:2488 weakly excites
- L27:1630 weakly excites
- L15:1003 weakly suppresses
- L18:2285 weakly suppresses
- L20:1580 weakly suppresses
- L15:1940 weakly suppresses
- L14:1677 weakly suppresses
- L11:2414 weakly suppresses
- L12:2165 weakly suppresses
- L11:2215 weakly suppresses closing quote marks at the end of speech
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
- 风俗
- '/');⏎
- undy
- ()];⏎
- lessly
- JSGlobal
- IGATION
- '/';⏎
- 祖国
- Subview
Suppresses
- 她说
- said
- ices
- Big
- 他说
- "
- said
- _comment
- istr
- Talk