'or' in ranges and alternatives
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. 21 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 reacts to the word 'or' when it joins two options or a range, such as 'age 14 or 15' or 'an hour or less'. Turning it up did not produce anything clearly about that, just wandering text.
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.
-
has found that most studies on college hunger reveal food insecurity rates among U.S. college students to be higher than the national average for adults not attending a two- or four-year institution.⏎When students are food-insecure, they are more likely to
-
Its greenhouse gas effect is huge with over twenty thousand times the effect of carbon dioxide over a hundred years and remaining in the atmosphere for a millennium or so.⏎In spite of these alarming figures its actual contribution at current levels of production
-
sensor on a piece of paper towel coated with water-resistant substance all over, except the areas showing four blood types and a positive or negative Rhesus factor. These areas are, packed with antibodies, interact with
-
this method will help you achieve a bouncing ball animation in probably 5 minutes or less.⏎So without wasting much time, let’s get into it!⏎Here are
-
Iowa. A lot of kids work for neighbors, and there are certain things parents would like their kids to be able to do when they reach age 14 or 15—including running a tractor and other farm machinery, Lang says.⏎The
-
MIT is one of the best universities in the world. Are you able to solve the problem that supposedly took an MIT professor a long time to solve in a day or less? Find out!⏎How can you spend your Math Minutes this week?⏎-
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: for me, the first time I will be 1000 times more often going to be: (1000) * (1000) * (1000) * (1000)
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
While not providing enough of certain minerals is a concern, oversupplementation can also cause problems. Pet food manufacturers keep a close eye on just how much of each and every mineral is used in a dog food. Minerals used must also be able to
-
971 and 9.8 million in 2001, according to a report published in the March 11 issue of Morbidity and Mortality Weekly Report.⏎And the number of cancer survivors continues to grow because of
-
- 3 Pulse transformer⏎- 4 RF transformer⏎- 5 Audio transformer⏎- 6 Other types⏎- 7 See also⏎- 8 References⏎This is the most common type of transformer, widely used in electric power
-
keywords for each book and chapter.⏎If you think you should have access to this title, please contact your librarian.⏎To troubleshoot, please check our FAQs, and if you can't find the answer there, please contact us.<|endoftext|>Indigenous
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
- L14:801 excites 'and' or 'or' joining list items
- L15:727 excites
- L12:51 excites
- L9:92 suppresses "and" or "or" joining two items
- L17:3025 suppresses
- L0:313 excites "and" joining paired nouns
- L6:427 suppresses
- L4:57 suppresses
- L5:1469 suppresses
- L4:108 excites "or" offering a choice in web text
- L18:1386 excites unclear: words before 'and' or 'or'
- L9:2010 suppresses
- L13:131 excites 'and' joining parallel phrases
- L3:250 excites
- L15:217 suppresses
- L8:70 excites 'and' between stacked superlatives
- L16:376 suppresses 'or' joining alternatives
- L9:77 excites commas in citations and name lists
- L12:1235 suppresses commas inside numbers and lists
- L18:2197 suppresses
- L7:999 suppresses
- L16:2940 excites commas inside lists and names
- L11:682 excites gap before a second number
- L1:1250 suppresses
Feeds later layers
- L25:1694 weakly excites 'and' joining paired names or fields
- L22:198 weakly excites
- L26:1007 weakly excites
- L26:339 weakly suppresses
- L24:2406 weakly suppresses
- L27:1163 weakly suppresses
- L27:753 weakly suppresses
- L27:909 weakly excites
- L26:1997 weakly suppresses
- L26:62 weakly excites
- L24:80 weakly suppresses
- L27:2267 weakly suppresses
- L24:1977 weakly suppresses
- L27:743 weakly suppresses
- L26:274 weakly suppresses
- L25:446 weakly excites
- L24:2151 weakly suppresses
- L27:1233 weakly suppresses
- L25:2770 weakly suppresses
- L23:1172 weakly excites
- L23:1445 weakly excites
- L24:1951 weakly suppresses
- L22:767 weakly suppresses
- L27:2632 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
- そうで
- /or
- eler
- IOR
- тому
- �
- orra
- ivers
- =post
- oin
Suppresses
- ируем
- 第一步
- ニア
- ву
- 的第一个
- 第一位
- 第一天
- nx
- CompatActivity
- .'));⏎