'and' or 'or' in split hyphen compounds
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. 14 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
The neuron lights up on the word 'and' or 'or' when it comes right after a hyphenated prefix, as in 'short-and long-term' or 'mis-and dis-information'. The steering test gave rambling, unrelated text, so it does not show a clear effect on 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.
-
curb would be difficult, if not impossible to do without the ability to maintain single limb support for a given amount of time. The ability to switch from two- to one-leg standing is required to perform turns, climb stairs and dress.⏎As we
-
of text. (Buzzfeed produced a helpful catalogue of all the fabricated and misleading stories distributed online.)⏎Visuals are regularly being used as persuasive vehicles for mis- and dis-information. And photographs are not the only kind of visuals to consider. Graph
-
getting organized. It offers tips for setting short- and long-term goals, choosing activities and fitting them into your daily life, and managing
-
several days, foraging movements confirmed by simultaneous acoustic observations of prey-sized fauna, movements parallel to the shore, and traveling between FADs. This intra- and inter-individual variety of behaviour might depend on the
-
the amount of random error involved in measuring student performance. The risk is that teachers and schools may be wrongfully rewarded or punished because value-added techniques either over- or underestimated their students’ learning gains. However, no intelligent users of standardized testing would
-
autopsies at 45 months. The effects of SIP forms on micro- and macrocell corrosion of reinforcement were investigated by statistically comparing the performance of specimens with and
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 have a 2000000000000000000001, which is 1,0000000000000000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
||Environmental management|⏎Class 10 Science Chapter 4 Environmental management questions answers in Pdf format:⏎Here is what you all are waiting for. Teachers, parents and student, now you will find below the Solutions of Class 10 Science Chapter
-
⏎Lin CC, Lin XM, Zou Y, Zhang L (2005) Study on the relationship between meteorological conditions
-
virus associated with multiple sclerosis⏎Science | January 13, 2022⏎Bjornevik K, Cortese M, Healy BC, Kuhle J, Mina MJ, Leng Y, Elledge SJ
-
your medicine is not helping or if you have side effects. Tell him or her if you are allergic to any medicine. Keep a list of the medicines, vitamins, and herbs you take. Include the amounts, and when and why you take them
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
- L10:2046 suppresses
- L7:1382 suppresses
- L7:616 excites
- L10:338 weakly suppresses
- L13:1548 weakly excites
- L9:2244 weakly excites
- L13:615 weakly suppresses
- L2:747 weakly suppresses
- L4:1846 weakly suppresses
- L12:372 weakly suppresses
- L13:876 weakly excites
- L9:92 weakly suppresses "and" or "or" joining two items
- L1:1678 weakly excites
- L7:737 weakly suppresses comma after a place name
- L11:807 weakly suppresses
- L9:1449 weakly suppresses
- L9:940 weakly suppresses
- L7:391 weakly excites 'of' after 'those' or 'many'
- L7:2996 weakly excites
- L3:182 weakly suppresses
- L13:499 weakly suppresses
- L12:2458 weakly suppresses
- L7:173 weakly excites
- L7:2588 weakly excites
Feeds later layers
- L15:2239 suppresses
- L15:1475 suppresses
- L15:1230 suppresses
- L16:2199 excites
- L18:745 weakly excites
- L15:1653 weakly excites
- L19:1092 weakly excites
- L15:631 weakly suppresses
- L15:2642 weakly suppresses
- L26:1257 weakly excites
- L26:112 weakly suppresses
- L15:1260 weakly suppresses
- L16:1812 weakly suppresses
- L20:2262 weakly suppresses
- L18:2278 weakly excites
- L24:748 weakly excites
- L15:3057 weakly excites
- L15:2923 weakly excites
- L18:573 weakly suppresses
- L15:2169 weakly excites
- L25:1673 weakly excites
- L24:407 weakly excites
- L17:1400 weakly excites
- L16:2531 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
- ulet
- otive
- füg
- ью
- ometimes
- وس
- нд
- rooms
- icial
- שות
Suppresses
- waves
- 支持力度
- ante
- =”
- 性疾病
- “At
- ância
- _HandleTypeDef
- anza
- akukan