unclear: mid-sentence function words
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. 20 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 is hard to pin down. It lights up on small connecting words and full stops in ordinary sentences, often right before the next phrase starts. When it was turned up, the model got stuck repeating a few words.
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.
-
and distributed to classrooms all over the world. They also offer workshops to assist K-12 teachers interested in integrating bioenergy into their curricula. In addition, a bioenergy literacy search tool is available to help education professionals find
-
that your data is backed up and accessible even if one data centre experiences an outage or disruption. Regular data backups protect against data loss and provide continuity for your business.⏎Cloud providers have dedicated teams of security experts who continually monitor and update security protocols.
-
physics professor Wilhelm Röntgen who discovered the use of an X-Ray way back in 1895.⏎Google doesn’t forget anyone though, and to celebrate, visiting their homepage will bring up a cool blue neon design. You may
-
so famous that it has been used as the title of books, movies, and albums.⏎It has also been used to describe periods of history.⏎The reason for this line’s popularity is quite understandable – many people have experienced a season of discontent.⏎
-
his own queen and provided her with a pension instead. A year earlier she had taken out a lease on the mansion of Cheynegates in Westminster Abbey but by 1492 she was resident at Bermondsey Abbey where she made
-
2.054 of the Texas Property Code governs judicial<|endoftext|>Adjustment to college life after school is always a challenge for young adults. One reason to be unsuccessful in navigating stress, depression, anxiety during college days is that the maturity 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: the following is a well-known example of a well- for which the following is so well-known, and so it can be generalized, but I can't for that, so it is on the following I am with that, and so I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
earnings.⏎Preferred stock, or preference shares.⏎To appeal to investors who wish to be sure of receiving dividends regularly, many companies issue what is called preferred stock, or preference shares. This class of stock has a prior claim to dividends paid by the
-
situation.⏎Sometimes you might want to return objects of different types based on arguments to the function. For instance, did the caller ask for the data in this format or that format?⏎Sometimes you'll want all of these variations at the same time.
-
like that (although, of course, the weights don't weigh anything) by creating a virtual reality simulator called the Cave Automatic Virtual Environment, or just CAVE, that works by tracking your muscle movements. Users put on a pair of goggles and
-
from Greenwich, England. Because of this strategic position, the humble outcrop of coral and sand became an important point in the journey by sea and later, air, between the US and Asia. The first attempt to establish Midway as a strategic
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
- L19:2215 excites
- L19:1516 excites
- L18:569 suppresses
- L19:223 suppresses
- L14:177 excites
- L19:1875 excites
- L19:2793 suppresses
- L19:413 suppresses
- L19:535 excites
- L19:1713 excites
- L18:1479 suppresses
- L9:11 excites
- L19:1554 excites
- L19:2037 excites
- L19:1659 suppresses
- L16:2531 suppresses
- L18:93 suppresses
- L17:336 suppresses
- L14:67 excites
- L11:29 suppresses
- L19:1662 excites
- L18:116 suppresses
- L19:840 excites
- L13:81 weakly excites
Feeds later layers
- L27:239 strongly excites
- L24:563 strongly suppresses
- L27:467 strongly suppresses
- L25:477 suppresses
- L21:720 suppresses
- L26:187 excites
- L23:100 suppresses comma or full stop ending a clause
- L21:102 excites
- L23:598 excites
- L26:836 suppresses
- L27:1300 excites
- L21:177 suppresses
- L26:11 suppresses
- L27:775 suppresses
- L21:717 excites
- L26:993 excites
- L23:235 suppresses
- L25:452 suppresses
- L25:469 suppresses
- L25:334 suppresses
- L21:930 suppresses
- L21:344 excites
- L26:255 excites
- L27:575 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
- .").
- .");
- .";
- .",
- ."));⏎
- )。
- .");⏎
- %).
- .');
- ?",
Suppresses
- )
- \)
- ()
- )
- ')'
- }
- ]
- ()
- ""
- ''