unclear: sentence ends in how-to text
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. 15 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 tends to fire on the full stop or line break that closes a sentence in advice, instructions or educational text. It also fires on a few unrelated spots, and its negative side seems to dislike the subject and verb in grammar examples about passive voice. Turning it up made the model get stuck repeating words, and turning it down gave plainer prose, so its job is hard to pin down.
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.
-
. Its high fiber content will also help you feel fuller with fewer calories. Try adding broccoli to your Asian-inspired stir-fries for a quick, nutrient-packed meal.⏎One last note: buy organic produce when you can to avoid
-
solution A) isotonic B) hypertonic C) hypotonic D) none of these I think it's B. I'm not too sure though.⏎November 27, 2007<|endoftext|>Hopefully your child will not experience
-
in San Mateo County compared to previous years; however, ticks are present and the diseases they carry can be very serious. Learn how to prevent tick bites here.⏎Click on the tick below to see a video of a tick 'quest
-
denominators or how to regroup or to borrow with mixed numbers? Take a deep breath and let us take it one step at a time. Click and begin.⏎GCF and LCM by Prime Factorization⏎A straight forward way to find
-
honesty⏎Start early to show your child that you trust him and that honesty is important to you, and you'll lay the groundwork for rearing an honest adult.⏎How to teach your child responsibility⏎To instill a sense of responsibility in your
-
2017 | Max-Planck-Institut für Bildungsforschung⏎A sudden drop in outdoor temperature increases the risk of respiratory infections⏎11.0
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 want to, as a computer, make this part of the world, which is my computer, to make it more efficiently and, as a computer, to use some other parts, which are available in the world, to develop some other
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
by Rita. Future (or modal verbs) : A letter will be written by Rita.⏎Example of Present Simple: Active: Tom cleans the house. Passive: The house is cleaned by Tom. Example of Past
-
to describe actions that began and were completed in the past. E.g. I had gone to the shop and bought some chips.⏎Other examples include.⏎After she had read it from start to finish, Sophie ripped the letter in half.⏎The frog
-
dog killed a ratA rat was killed by the dog The teacher will explain the story The story will be explained by the teacher⏎Examples Simple present: A letter is written by Rita. Simple past: A letter was written by Rita. Present perfect
-
geyser and found that the distribution is symmetric. Draw a boxplot of these data. Use the boxplot and quartiles to ...The following data represent the ages of chief justices of the U.S. Supreme Court when they were appointed.
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
- L13:2182 suppresses
- L14:1257 excites
- L13:2363 suppresses
- L13:1154 suppresses
- L13:1860 excites
- L12:215 excites
- L13:2454 suppresses
- L13:1407 excites
- L13:1620 suppresses
- L14:2434 excites
- L14:993 suppresses
- L13:646 excites
- L13:381 excites
- L13:51 excites
- L14:1117 suppresses
- L11:1567 excites
- L13:680 excites
- L12:757 excites
- L14:2631 excites
- L14:1828 weakly suppresses
- L14:1125 weakly excites
- L14:1569 weakly suppresses
- L14:2143 weakly suppresses
- L12:1932 weakly excites
Feeds later layers
- L16:1132 strongly excites
- L16:2525 excites
- L18:497 excites
- L16:1733 excites
- L16:2916 suppresses
- L16:1498 suppresses
- L16:1904 excites
- L16:1034 excites
- L16:510 suppresses
- L20:204 excites
- L17:251 excites
- L17:2726 excites
- L16:1445 excites
- L16:613 weakly excites
- L16:1688 weakly excites
- L17:2629 weakly excites
- L18:379 weakly excites
- L16:767 weakly suppresses
- L18:664 weakly excites
- L16:330 weakly excites
- L16:472 weakly excites
- L16:2717 weakly suppresses
- L16:1227 weakly excites
- L16:962 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
- another
- 又
- fox
- pac
- Another
- ito
- ebenfalls
- 璧
- 又有
- issa
Suppresses
- rine
- ucks
- iences
- atisation
- ulist
- ophysical
- recru
- plevel
- -scripts
- aises