'here' in 'read more here' links
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. 11 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 lights up on the word 'here' in phrases like 'read more here' or 'check it out here', which point the reader to a link. Turning it up made the output start filling with web links and URLs.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
This is considered a simple problem, one that a fifth grader may find in a math book. What about the problems we face in adulthood? (read more here)<|endoftext|>Задание от гостя:Размещено 1
-
illustrative case studies covering an array of disease categories, together with patients’ eGFR readings.⏎Note: Only Diamond member can download this ebook. Learn more here!<|endoftext|>March 28, 2013⏎Scientists Turn To
-
combined sewer systems. Additionally, this bill will establish a Green Infrastructure Program housed within the EPA's Office of Water. The full text of this bill can be read here. Participate in the Surfrider's action alert and send your Representative a
-
to recording video for upcoming movie shorts about the basin. Read more here.⏎The U.S. Fish and Wildlife Service concluded the sicklefin redhorse
-
journalistic institution, Adafruit strives to be a fair, informative, and positive voice within the community – check it out here: adafruit.com/editorialstandards⏎Stop breadboarding and soldering –
-
that the most effective way to intervene is also the easiest (not to mention the most time-effective.) Use brief messages, such as:⏎Check out the full article here:⏎“Get to know them, indulge your curiosity, spend time learning about who
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: the last time I was thinking about, the last time I was thinking about, the last time I was thinking about. So we are not the last time, we are not the time, we are not, the last time, we,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
00 wounded.⏎- Richter, G (1870). Manual of Coorg. Stolz. p. 337. Retrieved 14 August 2014.⏎- Archer, Jeremy. "Sir
-
||L Education > L Education (General)|⏎|Deposited By:||Zaidatun Tasir|⏎|Deposited On:||23 Jul 2008 0
-
-Earth objects⏎- Interactive: Close encounters of the asteroid kind⏎Connect with the Cosmic Log community by "liking" the log's Facebook page, following @b0yle on Twitter or following the Cosmic Log Google+ page. You can
-
) Swarm of T-wave events during October 2005⏎Please cite this report as:⏎Global Volcanism Program, 2007. Report on Macdonald (Undersea Features) (Wunder
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
- L10:2158 excites
- L7:2492 excites
- L10:2802 excites
- L3:1836 suppresses 'the' after 'lack' or 'have'
- L8:1291 excites
- L8:1616 weakly excites
- L8:1271 weakly excites
- L10:1107 weakly suppresses
- L6:959 weakly excites
- L8:950 weakly suppresses
- L9:2779 weakly excites
- L10:615 weakly suppresses
- L6:1094 weakly suppresses
- L10:439 weakly suppresses
- L6:1359 weakly excites
- L9:2871 weakly suppresses
- L7:10 weakly suppresses
- L9:2327 weakly suppresses
- L7:666 weakly suppresses
- L10:2852 weakly excites
- L10:1251 weakly excites
- L9:291 weakly suppresses
- L5:2406 weakly suppresses
- L10:349 weakly suppresses
Feeds later layers
- L17:1964 excites
- L22:1354 excites
- L21:2436 excites
- L19:943 excites
- L15:1066 excites
- L21:598 excites
- L13:1710 excites
- L19:2961 excites
- L14:2474 excites
- L15:2752 excites
- L21:2543 excites
- L13:2821 weakly excites
- L19:2438 weakly excites
- L16:2539 weakly suppresses
- L12:1288 weakly excites
- L21:1696 weakly suppresses
- L20:1381 weakly excites
- L26:600 weakly excites
- L22:2352 weakly excites
- L14:1012 weakly excites
- L17:1714 weakly excites
- L26:497 weakly excites
- L17:2408 weakly suppresses
- L18:947 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
- tutorials
- Encyclopedia
- 请点击
- 全文
- https
- 详见
- articles
- article
- 我就
- reviews
Suppresses
- -West
- tright
- struments
- urm
- inded
- pline
- heck
- 的速度
- strument
- 褥