"make" or "do" before an object noun
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. 4 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 fires on the verb right before a phrase like "make a difference", "make sense", "make a list" or "do some research". When it was turned up, the writing drifted into vague, general talk about everyday topics.
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.
-
the long-term benefits it can provide. It's vital to make a difference and discover your life's purpose at any age and in any circumstance.⏎Vol
-
wisdom to the young. This sharing of information is a way to nurture and protect our children as it empowers them to make sense of the past, to make wise decisions in the present, and to cultivate a
-
point of view and brainstorm for solutions. Put yourself in the other person’s shoes (empathy - the ability to feel with -)⏎5. Never make a complaint or air a griev
-
a weakness in your infrastructure that could potentially expose you to threats.⏎Measuring Organizational Risk⏎The first step to measuring risk is identifying threats. Start by making a list of threats to your organization’s digital assets. From there,
-
it’s time to do some serious watering.⏎If your succulent is outdoor, it’s a good idea to
-
afford it, you can try it on your own. Don’t embark on any new diet restrictions without a doctor’s supervision if you are ill, pregnant or breastfeeding
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 a research project to do, and I have been the effort of this statement. I have been the effort of this statement. I have been the effort of this statement. I have been the effort of this statement. I have
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the analyst, the developer and the tester – have a common understanding of the scope of the work. From this they can agree a common definiton of “done
-
prolong the cycle of life, human or non. Sustainable decisions and actions are those that cause no harm and consider the long term effects.⏎So depending on the entire scope of a project’s conception and construction, a building with green parts isn't necessarily
-
website.<|endoftext|>After reading this chapter you will…⏎- understand graphs as a mathematical structure.⏎- be able to traverse graphs using well-known algorithms.⏎- learn the scope of problems that can be addressed using graphs.⏎Graphs are perhaps the most versatile
-
outside by identifying business outcomes, and then drills down into the feature set that will achieve those outcomes. Each feature is captured as a “story”, which defines the scope of the feature along with its acceptance criteria. This article introduces the BDD approach
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
- L2:2692 excites
- L3:50 weakly excites
- L3:456 weakly suppresses
- L0:194 weakly suppresses
- L2:1710 weakly suppresses
- L2:111 weakly suppresses
- L3:1247 weakly suppresses
- L3:2326 weakly excites
- L3:464 weakly excites
- L1:342 weakly excites
- L0:467 weakly suppresses
- L1:3017 weakly excites
- L1:828 weakly suppresses
- L2:49 weakly suppresses
- L2:2997 weakly excites
- L3:421 weakly suppresses
- L1:2891 weakly suppresses
- L1:2380 weakly excites
- L3:1756 weakly suppresses
- L3:2639 weakly suppresses
- L3:1266 weakly excites
- L2:1608 weakly suppresses
- L1:1593 weakly suppresses
- L0:1784 weakly suppresses
Feeds later layers
- L5:2825 weakly excites
- L5:2552 weakly suppresses
- L6:1556 weakly suppresses
- L8:2897 weakly suppresses
- L19:2132 weakly suppresses
- L16:2683 weakly excites
- L8:2797 weakly suppresses
- L10:299 weakly suppresses
- L9:1015 weakly excites
- L12:911 weakly suppresses
- L10:2430 weakly suppresses
- L16:1642 weakly suppresses
- L22:1196 weakly suppresses
- L22:2511 weakly excites
- L7:2179 weakly excites
- L22:2856 weakly suppresses
- L7:2292 weakly suppresses
- L26:2859 weakly suppresses
- L9:892 weakly suppresses
- L6:2425 weakly suppresses
- L13:2402 weakly suppresses
- L16:1297 weakly excites unclear: document ends and answer colons
- L12:213 weakly excites
- L24:2779 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
- ложения
- aments
- ))/(
- pdata
- eos
- amente
- ся
- 直至
- junto
- -rating
Suppresses
- ��
- Scope
- scope
- şık
- SCO
- ła
- OfWork
- Scope
- otre
- .Scope