"how do you" question phrase
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. 0 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 "you" right after "how do" in questions, such as "How do you outline notes?" Turning it up made the model write questions and prompts, which fits the trigger.
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.
-
headings.⏎How do you outline notes?⏎drawn, the outline method should be replaced with a better note
-
use the networks for communication purposes. In education, we have online libraries that we can visit on our home PC. All of this is<|endoftext|>Describe a how do you cite a research paper time when you had to solve a problem without classification essay writing
-
imagination in motion.⏎How do you send your child to NERF Camp?⏎All you have to do is to
-
's disease, and how do you differentiate between regular memory loss and memory loss associated with Alzheimer's disease?⏎Answer:
-
an edited transcript.⏎Mainebiz: How do you think increasing commercial demand for forest products will affect Maine's forests?⏎Barton:
-
polymorphisms that can be analyzed in a human nuclear background, and provide a novel model for the study of nuclear-mitochondrial interactions.<|endoftext|>How do you learn best?⏎In a recent post, I mentioned the concept of learning styles.
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 thinking about how to use the information to make a question for you to answer. So, for example, if you are thinking about how to use the information to make a question for me to answer, then I can say:
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
thraquinone, which has a natural chemical attraction<|endoftext|>From: Michael Barrie <mibarrieinterchange.ubc.ca>⏎Subject: Overt Markers on Mass Nouns⏎E-mail this message to a friend⏎Query
-
differing modes of inheritance. Agreement appears to be near at hand that IDDM and NIDDM exhibit substantial genetic differences.⏎Stages of Diabetes⏎1. Overt or manifest diabetes⏎It refers to the fully expressed clinical syndrome. This stage of
-
Columbus⏎The long low ship, With a high stern and pierced for many oars, glides effortlessly up the long narrow bay under its towering square sail.⏎After
-
days, replenishing the ship with food, water, precious exotic hides, and many other things, the got underway hours before the morning breeze. Their many oars glided their craft like a ghostly apparition southward down the bay toward
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
None: this is the first layer.
Feeds later layers
- L8:474 weakly excites
- L23:1016 weakly excites
- L25:3016 weakly excites
- L7:1409 weakly excites
- L14:1255 weakly suppresses
- L11:825 weakly suppresses
- L3:3038 weakly suppresses
- L3:1879 weakly suppresses
- L16:2696 weakly excites
- L12:2204 weakly excites
- L23:1444 weakly suppresses
- L17:741 weakly excites
- L1:1979 weakly excites
- L5:2315 weakly excites
- L11:2223 weakly excites
- L22:1105 weakly excites
- L19:1095 weakly suppresses
- L4:595 weakly suppresses
- L7:73 weakly suppresses
- L15:2469 weakly excites
- L15:1507 weakly suppresses
- L2:3034 weakly suppresses
- L20:1017 weakly suppresses
- L13:1622 weakly 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
- щен
- roid
- roids
- enaries
- hood
- loid
- hud
- 肌
- handled
- ryo
Suppresses
- DataTypes
- ')[
- 罢了
- 的人来说
- imate
- 邻居
- ':⏎⏎
- 这样说
- 靛
- ")[