'and' between stacked superlatives
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. 8 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 'and' when it joins two describing words, like 'the oldest and most organized' or 'the newest and best'. It is quiet on 'as well as', which is the opposite pattern. Turning it up did not clearly change 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.
-
and the YMCA and the Salvation Army participated. In all, 130 U.S. agencies involved themselves in war relief. The ARC, the oldest and most organized relief group in 1914, had less than 1
-
class on code.org, jump into CircuitPython to learn Python and hardware together, TinyGO, or even use the Arduino IDE. Circuit Playground Express is the newest and best Circuit
-
on the deep web before becoming a popular commodity.⏎Most Popular Cryptocurrencies⏎Let’s take a look at the various cryptos on the market:⏎The oldest and arguably most successful cryptocurrency on the market is beyond a shadow of a doubt Bitcoin.
-
4th of July should be a time when we come together to honor our country by celebrating our great nation safely and responsibly," Hugh Parry, president and CEO of Prevent Blindness America, said in an organization news release. "We
-
counselor will play an increasingly important role as patients and doctors struggle with the impact of these exciting new breakthroughs. The Human Genetics Program at Sarah Lawrence College the oldest and largest training program for genetic counselors in the United States, is well positioned to take
-
Amazon.com.<|endoftext|>The world’s seabeds are littered with wrecks and treasures. Despite many decades of increasingly active searches for the best, the rarest and the most valuable of them
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 number of people who have had to take a break from the top of the pyramid. the number of people who have had to take a break from the top of the pyramid...? the number of people who have taken a break...?
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
effective data security, and a host of management features.⏎In this blog, we discuss how adopting an Integrated Learning System can help teachers maximize their own productivity as well as that of their students.⏎Monitoring student progress⏎Most integrated learning systems contain a host
-
the piezoelectric transducers point to their applicability to wide range of dynamic pressure field measurements.⏎Subject Headings: Cavitation | Dynamic pressure | Pressure measurement | Piezoelectricity | Transient response |
-
sweet-briar, damask, Cherokee, moss, monthly, musk, and "wild roses."⏎Beloved for centuries for their practical uses as well as their beauty, Old World roses were brought to the New World by European settlers who
-
the Day, Vol. VII, pp. 487-489.⏎IT is a pleasure to speak before this association whose service to business, to agriculture and to labor of these many years is so established and so esteemed. 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
- L4:57 suppresses
- L6:912 suppresses
- L6:427 suppresses
- L7:42 excites
- L7:999 suppresses
- L1:1250 suppresses
- L6:1044 suppresses comma in lists before "and" or "nor"
- L7:23 suppresses
- L0:313 weakly excites "and" joining paired nouns
- L3:71 weakly suppresses
- L6:2945 weakly suppresses
- L0:68 weakly suppresses
- L6:2534 weakly suppresses
- L6:2682 weakly excites
- L3:1951 weakly suppresses
- L6:2466 weakly suppresses
- L6:244 weakly suppresses
- L5:1242 weakly suppresses
- L7:632 weakly excites
- L3:250 weakly excites
- L3:1725 weakly excites
- L6:1108 weakly suppresses
- L6:1121 weakly excites 'while' joining two clauses after a comma
- L3:274 weakly excites
Feeds later layers
- L12:51 strongly suppresses
- L13:125 excites
- L10:144 excites
- L16:1659 excites
- L11:251 excites
- L9:212 suppresses
- L14:1054 excites
- L9:92 excites "and" or "or" joining two items
- L16:1294 excites
- L15:677 excites
- L14:1639 excites "and" joining two nouns
- L15:545 excites
- L13:371 suppresses
- L14:874 excites
- L12:1212 excites
- L12:1786 excites
- L13:716 excites
- L19:1823 excites 'and' inside organisation or paired names
- L17:1499 excites
- L12:2058 excites
- L11:682 excites gap before a second number
- L15:217 suppresses
- L17:3061 excites
- L15:1162 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
- /OR
- /or
- 맞
- ulant
- ulkan
- ulate
- ULATOR
- currentPage
- cudaMemcpy
- /
Suppresses
- paring
- itian
- лон
- it
- 披
- 宁愿
- on
- etched
- 的不同
- 那样的