'not' in 'but not limited to'
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. 16 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 mostly fires on the word 'not' in the phrase 'but not limited to', which is a common way to say a list is incomplete. It also fires on a few unrelated words like 'Credit' and 'Protect'. Turning it up did not produce a clear theme in the 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.
-
a common food source for many Native American tribes including but not limited to the Iroquois, Algonquin, Ojibwa
-
documents for those who are going to build it for you and also for legal purposes. We will cover all these things in details. Some of the topics include but not limited to:⏎- How to Plan and Draw/design Your Dream Home⏎- Understanding
-
the world’s rainforests and that frontline communities organizing against the extraction and burning of dirty fossil fuels deserve the strongest support we can offer. RAN established the Protect-an-Acre program to protect the world’s forests and the rights of their inhabitants
-
and pesticides, extensive soil erosion, pollution, and high rates of greenhouse gas emissions. Low-intensity farming methods on the other hand–which include, but aren’t limited to, organic agriculture–typically favour more environmentally-friendly farming practices at the expense
-
domains, known as the acceptor and the donor layers. Excitons, the energy particles created by solar cells, must be able to travel quickly
-
repeat it at any time. Just add a new Alka-Seltzer tablet and watch as the colored bubbles rise again!⏎2. Lemon Volcano⏎Image Credit: sixthbloom.com⏎Lemon volcanoes don’t create a gigantic eruption
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 a 1 1 1, and I have a 1, one, one, one, one, ONE, that is, that is, that is, that is, that is, that that is, that is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
production.2⏎That support has paid off: the state is one of the nation’s leading producers of biofuels, pumping out over 1.2 billion gallons per year.3⏎However, all is not well on the policy front
-
a site on Crete known for its remarkably cool and stable conditions.⏎They recovered DNA in teeth from 32 individuals who died between 2290<|endoftext|>Module 7: Molecular Computing with RNA Sequences⏎Intron/Exon
-
accounted for the population of adults who reported mobility issues in the NTS).⏎So, car trips for 1 mile or under add up to 1.8 billion miles a year… and that’s just in England.⏎These short journeys
-
warming _ precious water. Officials say the system could end up saving 20 percent of the water Spain uses for irrigation today _ a whopping 1.3 trillion gallons of water per year.⏎It is a major milestone for a country that is
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
- L15:12 strongly suppresses
- L8:20 excites
- L10:50 suppresses
- L13:33 suppresses
- L8:7 suppresses
- L8:4 suppresses
- L6:13 suppresses
- L9:13 suppresses
- L6:19 excites second half of proper names
- L6:7 excites
- L6:16 suppresses
- L6:12 suppresses
- L6:8 excites
- L0:1224 excites
- L8:14 suppresses
- L4:133 suppresses
- L12:713 excites
- L8:10 suppresses
- L1:2683 suppresses
- L5:198 suppresses unclear: weak "and" in lists
- L1:951 suppresses
- L8:16 excites
- L5:50 suppresses
- L5:102 suppresses
Feeds later layers
- L22:332 excites
- L21:110 excites
- L25:182 excites
- L25:854 suppresses
- L24:371 excites
- L19:418 excites
- L19:1057 suppresses
- L24:8 excites
- L19:258 excites
- L26:1451 excites
- L19:785 excites
- L20:761 excites
- L19:311 suppresses
- L19:848 suppresses
- L19:43 suppresses
- L17:1409 suppresses
- L17:529 suppresses
- L19:2850 suppresses
- L27:428 suppresses
- L18:1048 suppresses
- L18:2741 suppresses
- L19:814 suppresses
- L19:737 suppresses
- L19:2519 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
- (std
- 万
- static
- urrenc
- fbe
- lever
- тив
- overloaded
- ComVisible
- eparator
Suppresses
- ness
- 江区
- HEAP
- breakdown
- Cook
- sist
- �
- ien
- raq
- eref