unclear: qualifiers like 'only' and 'alone'
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 mixed: they peak on different tokens. 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 words that qualify or contrast a claim, such as 'only', 'alone', 'anyone' or 'while', in the middle of an argument. It is hard to pin down more precisely. Turning it up produced rambling text with no clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
2. No other industry can directly impact local economies to the same extent as manufacturing.⏎Women’s inclusion in the manufacturing industry is pivotal to sustaining this success. Women only hold roughly
-
their designs, software and tutorials online and calling it a day: Just because something is available in theory doesn’t automatically make it accessible to all. (Linux, anyone?)⏎Like a whole cadre of other prosthetic researchers and makers, they are interested
-
Establishing a program of incentives and penalties is not enough, alone, however. Enforceability is important—the market alone can’t make improvements permanent. Questions also remain as to how long the greenhouse gas
-
users from accessing “valid” content (blocking them from using their devices) and also opened up security holes.⏎Intellectual property should be respected (the digital world runs on it) but we should not do so at the expense of privacy or security
-
Unfortunately, due to our continued poisoning of the environment and the Fukushima power plant meltdown, many fish are now loaded with unsafe levels of mercury and radiation. All fish, every single fish on this planet, have some level of mercury; however, some
-
Mike Fortes to encourage countries to consider the values of seagrass when managing coastal environments.⏎This has been called the Bolinao Declaration and reads:⏎We, participants of the third International Seagrass Biology Workshop (ISBW 3),
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 been thinking about the different things, the different things, the different things. The different things, the different things, the different things, the different things, the different things, the different things, the different things, the different things
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Now, your shoes can not only count your steps but also study them.⏎“This means that now you can understand how your feet are being placed on the ground, how much pressure is being applied on the bones, the angle of your steps and more
-
of english⏎teachers is going to be english⏎enrichment rather than teaching english⏎to beginners. The ability to teach someone a language with its expressions, its metaphors and its cultural and historical context is not only a large proposition but also
-
and Testing. Though they’re closely related, they are, ultimately, different.⏎If you work in the IT industry, you’ve probably come across them. You’ve also probably noticed that many executives – and customers – don’t understand the difference between
-
your wheels because you don’t even know where to start.⏎I’ve said this before and I’ll say it again: Teens have SO much going on! So much earlier than we had to. College prep, varsity sports, volunteer work/community
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
- L13:1180 suppresses
- L10:1459 suppresses
- L13:1066 suppresses
- L12:982 excites
- L12:990 suppresses
- L12:690 excites
- L8:891 weakly excites
- L11:1251 weakly excites
- L13:557 weakly excites
- L10:1164 weakly excites
- L13:1897 weakly suppresses
- L10:782 weakly suppresses
- L10:2953 weakly excites
- L13:2710 weakly suppresses
- L10:713 weakly suppresses
- L12:107 weakly excites
- L4:22 weakly suppresses
- L3:1641 weakly suppresses
- L7:1656 weakly excites
- L13:1296 weakly excites
- L13:912 weakly suppresses
- L1:2954 weakly suppresses
- L13:1193 weakly excites
- L2:2900 weakly suppresses
Feeds later layers
- L16:2442 excites
- L16:2329 suppresses
- L17:1099 suppresses
- L15:270 suppresses
- L19:1605 suppresses
- L15:1949 suppresses
- L17:632 suppresses unclear: coordinating words in lists
- L16:1475 suppresses
- L15:2050 suppresses
- L17:919 suppresses
- L15:2862 suppresses
- L15:709 suppresses
- L15:732 suppresses
- L26:347 suppresses the 're' in self-reliant and similar
- L15:509 suppresses
- L25:6 excites
- L15:132 excites
- L16:1141 suppresses
- L15:2369 suppresses
- L15:1207 weakly suppresses
- L19:164 weakly excites
- L15:1504 weakly excites
- L15:1339 weakly suppresses
- L15:1681 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
- inction
- 报复
- 转载请
- '}}
- .SuspendLayout
- 唛
- betr
- iller
- '}}⏎
- ']]⏎
Suppresses
- and
- 、
- 和
- 和社会
- otional
- ,
- ~,
- icios
- 以及
- 和服务