"the" inside institution names
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. 15 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 word "the" when it sits inside the name of an organisation or journal, such as "Proceedings of the National Academy" or "Center for the Global Environment". The negative side shows up on hyphens and "of" in other spots, but those are weaker and less clear. Turning it up did not produce a clear theme in the model's writing.
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.
-
ational Sum Frequency Generation Spectroscopy as Probes of Bulk and Surface Compositions of Size-Resolved Sea Spray Aerosol Particles,” Phys. Chem. Chem. Phys., 2013, 15, 620
-
ons’ performance was on a par with that previously reported in baboons for this type of complex task. Their study, which is published in the journal Proceedings of the National Academy of Sciences (PNAS), is the first to identify a non-pr
-
people vigilant and careful,” said psychologist Jack van Honk of Utrecht University in the Netherlands, who led the study published May 24 in* Proceedings of the National Academy of Sciences*.⏎In the study, a few dozen females received half
-
culture to think of heat as risky,” said Bernstein, a pediatrician at Boston Children’s Hospital and co-director of the Center for Climate, Health, and the Global Environment at Harvard’s School of Public Health.⏎As global temperatures rise and extreme
-
<|endoftext|>Improved Screening for Adverse Childhood Experiences⏎The Institute for Innovation & Implementation at the University of Maryland School of Social Work has partnered with Harvard's Center for the Developing Child's Frontiers of Innovation Community to develop a new project aimed at better
-
Wageningen University and Research, the University of Groningen, Utrecht University, Vrije Universiteit Amsterdam, and the National Institute for Public Health and the Environment (RIVM) on research into
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: what is the, is the, (1) in, in, (1) in, () in, () in) in, () in, () in) in, () in) in) in) in
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
scale of 0-500, fourth-graders scored 224 and eighth-graders scored 274. The national average for fourth- and eighth-graders was 217 and 263, respectively
-
produced by the parotid glands is in nature, and enters the through the after passing through the which are prominent in the gland. Despite being the largest pair of glands, only approximately 25% of saliva is produced by the glands.⏎
-
thereof).Please provide resources as required by APA standards.⏎https://mynursingpaper.com/wp-content/uploads/2020/06/cropped-1559720319884139
-
.newworldencyclopedia.org/entry/Karl_Marx<|endoftext|>Credit: Andrei Marincas | Shutterstock
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
- L10:77 suppresses
- L9:7 excites
- L8:30 excites blank slot right before a number
- L9:1515 excites line break splitting a phrase
- L14:300 suppresses
- L4:1097 suppresses
- L6:0 excites
- L9:61 excites
- L10:29 suppresses
- L8:11 excites
- L5:5 suppresses digit after a label like paper or type
- L7:195 suppresses 'of' in set noun phrases
- L6:249 excites
- L14:348 suppresses
- L10:424 suppresses
- L11:31 suppresses
- L13:1046 suppresses 'in' or 'on' ending a phrase
- L14:493 excites
- L14:1018 excites
- L4:312 weakly excites 'that' after 'the way' or 'the first time'
- L14:915 weakly suppresses
- L4:513 weakly suppresses
- L13:132 weakly excites "of" after a noun
- L10:533 weakly excites
Feeds later layers
- L16:689 excites 'the' in fixed phrases like 'ahead of the curve'
- L16:337 excites
- L16:1258 excites
- L16:1003 excites
- L16:995 excites
- L27:262 excites
- L18:2887 excites
- L17:359 excites
- L16:992 excites
- L17:390 weakly suppresses
- L17:1555 weakly excites
- L19:373 weakly excites
- L17:146 weakly suppresses period after abbreviations and dates
- L16:1020 weakly excites
- L16:3040 weakly excites
- L17:137 weakly excites
- L16:2000 weakly excites
- L18:2189 weakly suppresses
- L20:2896 weakly suppresses
- L16:538 weakly excites
- L18:70 weakly excites
- L16:2652 weakly excites
- L20:1287 weakly excites
- L26:1941 weakly excites
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
- bilt
- tória
- venida
- ('{{
- ftware
- présent
- uition
- bulk
- 琬
- redients
Suppresses
- /object
- /head
- ingly
- 的那个
- /be
- amento
- delle
- отметить
- -ul
- /m