comma after a full clause
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. 19 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 a comma that comes right after a complete statement, where a new clause is about to start, like "struck by a distracted driver," or "emit a great amount of radiation,". When turned down, the model got stuck repeating "if" and "when".
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.
-
July, 2007. Lynn Grant and her daughters Shannon and Hannah were on the way to a birthday party when they were struck by a distracted driver, who ran a red light driving 55 mph. Hannah 6 years old
-
Thermostats regulate the system’s temperature, while filters clean the circulated air. Understanding these parts is crucial for effective maintenance and
-
As transistors, the building blocks of microprocessors, approach their minimum size limits, creative ways to continue increasing computing power have emerged. These include technologies such as pip
-
women starting at the age of 21, so she makes a mental note to ask specifically about any previous pap smears and enters the room to meet Stephanie, who is
-
It is the number of protons in the nucleus minus the number of electrons in between the nucleus and the electron in question. Basically, the nucleus attracts an electron, but other electrons in lower shells repel it (opposites attract, likes rep
-
diving real close to the side.<|endoftext|>Fashion and style are often used interchangeably, but they are two distinct concepts. While fashion refers to the trends and clothing items
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: by if to in when when when if if if if if if if if when when when when when if when when when when when if when when when when if when when when when if when when when when when if when when when when if
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. For years, researchers have sought to understand how such a small creature could make such a long journey, particularly because it entails crossing the Gulf of Mexico without stopping.⏎Stories told by generations past
-
’re giving each student a chance to consider the question, weigh arguments for and against each answer choice, and commit to what s/he thinks is the best answer–all before s/he hears what the other students think about the question.⏎All students are
-
. Often this is the only test done which is just not sufficient, especially if you are experiencing symptoms of low thyroid function.⏎What it does: TSH
-
the laws of relativity.⏎For example, the Official Handbook of the Marvel Universe states that the character Northstar maintains speeds which can be considered "modest," especially when carrying a passenger. The Handbook also concedes that a solid object moving in
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
- L17:3021 suppresses
- L14:35 suppresses
- L18:366 excites comma inside numbers and dates
- L18:93 suppresses
- L16:1059 excites
- L15:286 excites
- L16:846 excites comma opening an aside or explanation
- L16:797 suppresses
- L17:3068 excites comma after a sentence subject
- L4:3 excites commas after citation fields
- L16:240 weakly excites comma after 'sometimes' or 'the truth'
- L6:2362 weakly excites
- L18:2434 weakly suppresses
- L17:1597 weakly excites
- L11:154 weakly suppresses commas after citation numbers
- L5:16 weakly excites
- L17:1537 weakly excites
- L12:1295 weakly suppresses
- L17:585 weakly excites
- L11:2745 weakly suppresses
- L16:316 weakly suppresses
- L15:2835 weakly excites
- L15:630 weakly suppresses
- L17:5 weakly suppresses
Feeds later layers
- L24:148 strongly suppresses
- L20:458 strongly suppresses
- L21:140 excites
- L20:1245 suppresses
- L22:160 excites
- L21:1064 suppresses
- L23:449 suppresses
- L22:132 suppresses
- L24:116 excites
- L27:338 suppresses
- L23:756 suppresses
- L25:768 suppresses
- L22:478 suppresses
- L23:664 suppresses
- L26:852 suppresses
- L27:123 excites
- L25:626 suppresses
- L24:784 suppresses
- L27:511 suppresses
- L21:1597 suppresses
- L21:1023 suppresses
- L27:659 suppresses
- L26:509 suppresses
- L26:790 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
- ца
- 允许
- ,
- 拉
- 腾
- apest
- FAULT
- 半岛
- 是你
- ,t
Suppresses
- if
- when
- jika
- through
- via
- without
- during
- whenever
- when
- after