'off of' and 'of' after 'off'
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. 2 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 the word "of" when it comes right after "off", as in "feed off of blood" or "off of Broadway". Turning it up did not produce a clear shift in the model's 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.
-
as clothes, carpets, and, yes, beds, and they feed off of blood.⏎There are about ninety known species of bed bugs worldwide, but luckily for
-
Manitowoc Rapids. It is estimated that more than 5,000 people attended the event. Today, the marker is located off of Broadway Street.⏎In the days before settlement, our area was named Manitowoc
-
the world’s most fragile and endangered ecosystems. In the last few decades over 35 million acres of Coral Reefs have been obliterated. Reefs off of <|endoftext|>Time Ticks Away in Portugal⏎Compared with every other country in
-
red squirrels and porcupines.⏎Types of Flora⏎The animals of the park have plenty to forage and thrive off of with a strong botanical environment in Palo Verde National Park. Some of the plant species
-
using simple methods.⏎Now, caffeine has fairly good solubility in a variety of environments thanks to all those myriad polar and non-polar functional groups hanging off of it, which results in a wide range of possibilities for means of extraction, collection
-
group activities, toys, and other things the children will do within a week. Creating a newsletter for parents. I observed children and their development and assessed them off of what I observed. I also observed planning for field trips. I interacted with
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 working on some problems. I have been, I have been thinking about. experiences. the projects. this is. . The. The key, I can. will. is. online. will. activities in. course.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
9: Differentiate between www and internet.⏎problem 10: What do you mean by Usenet? List out main categories of it?⏎problem 11: Define
-
down, but cutting would only be done when the beetle is dormant to prevent further spreading, Eddy says. She adds, "There is no one textbook way of dealing with this because there can be so many different circumstances."⏎For now, the
-
methods you can use to overcome the blue light effects without cutting your TV watch time.⏎1. Protective Glasses:⏎Using protective glasses is one of the most effective ways of overcoming the problems of blue light. The glasses are available in the market that effectively
-
most successful people share the common trait of self-awareness. They recognize the situations that will make them successful, and this makes it easy for them to find ways of achieving objectives that resonate with their motivations.⏎They also understand their limitations and where they
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
- L1:2234 weakly suppresses
- L1:1129 weakly suppresses
- L1:1579 weakly suppresses 'Well' starting a reply or answer
- L1:2559 weakly excites
- L1:2945 weakly excites
- L0:1898 weakly suppresses
- L1:669 weakly excites
- L0:786 weakly suppresses
- L0:274 weakly excites
- L1:1152 weakly excites
- L1:1876 weakly suppresses
- L1:2268 weakly excites
- L1:535 weakly suppresses
- L1:2187 weakly excites
- L0:1651 weakly suppresses
- L1:1097 weakly excites
- L0:2093 weakly suppresses
- L0:1580 weakly excites
- L0:1499 weakly excites
- L1:1308 weakly excites
- L1:2620 weakly excites
- L1:149 weakly suppresses
- L1:2465 weakly excites
- L1:924 weakly excites
Feeds later layers
- L9:1782 suppresses
- L7:1157 excites
- L17:664 suppresses
- L8:1251 suppresses
- L8:514 suppresses
- L3:1525 suppresses
- L9:2913 excites
- L10:914 excites
- L10:664 suppresses
- L24:456 excites
- L24:1748 weakly suppresses
- L10:1042 weakly excites
- L14:979 weakly suppresses
- L9:1625 weakly suppresses
- L7:437 weakly suppresses
- L8:277 weakly excites
- L10:190 weakly suppresses
- L20:1637 weakly suppresses
- L26:843 weakly excites
- L16:692 weakly suppresses
- L7:887 weakly excites
- L24:824 weakly suppresses
- L23:3035 weakly suppresses
- L3:660 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
- -san
- erto
- [dir
- -driving
- cimal
- atron
- stered
- irement
- 发改委
- 指南
Suppresses
- /Error
- igham
- onga
- ึง
- /Core
- taş
- 岙
- ymce
- PAS
- og