research and study wording
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. 9 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 responds to words used when describing scientific studies, such as "law", "measures", "drug", "sensitivity" and "energy" in research reports. It goes strongly negative on health-claim phrasing like "linked to", "proven" and "backed up with research". Turning it up gave replies with little connection to studies.
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.
-
Medical Association.⏎The study was supported by the National Institutes of Health<|endoftext|>The generalized second law of thermodynamics states that entropy always increases when all event horizons are attributed with
-
) demonstrated that even low⏎levels of inulin are effective in decreasing plasma cholesterol in rats.⏎DESIGN: A prospective, within-subjects,⏎repeated measures design counterbalanced to prevent order effects will⏎be used. Small quantities of JA
-
, Herlitze's team was interested in a protein (G-protein-coupled receptor) which occurs in various cell types. Had the researchers administered a drug, they would not only have deactivated the receptor in the Purkinje cells,
-
of the generalized second law of thermodynamics. However, an intriguing set of open universe models shows an apparent
-
2011) have informed these revised guidelines. More information on HPV and cervical cancer.⏎DCEG research provided the initial assessment of the specificity, sensitivity, and appropriate applications of the first-generation HIV antibody testing
-
000 grant from the U.S. Department of Defense Breast Cancer Research Program to see if the answer to those questions is "yes". The study is investigating the very real possibility that dead cancer cells left over after chemotherapy spark cancer to spread to
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: "In a recent study, the effect of a single act of eating a piece of chocolate that contains 30% of the NMD, which is a compound that is commonly found in the brain's reward system. This study found that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
your living space, making your home the best place for relaxation.⏎Enhances Productivity and Creativity⏎Having plants either in your home or office has been linked to increased productivity and creativity. The greenery offers a vibrant energy that enhances creativity and
-
is a great time to form this habit as it will serve you well throughout your life.⏎For one, exercise increases one’s energy levels. It’s also been proven to replenish brain energy reserves, especially in areas of the brain responsible for cognitive
-
gratitude is different for men and women and the impacts this has on relationships, romantic or not. It provides practical examples of how to foster gratitude, backed up with research as to why this works and how to implement it.⏎If you’re looking for
-
your generosity.⏎The Benefits of Giving to Others⏎Generous Behavior Increases Happiness⏎Not only does giving to others increase their happiness, it also increases our own. Studies amongst different cultures and ages have consistently found that individuals who spend money on others report
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
- L7:1367 excites
- L4:756 excites
- L6:2309 suppresses
- L7:2362 excites
- L6:2851 suppresses
- L6:1786 excites
- L5:708 excites
- L7:435 excites
- L4:2715 excites
- L3:2420 excites
- L8:990 excites
- L6:1993 suppresses
- L8:2883 weakly excites
- L6:2532 weakly excites
- L6:222 weakly suppresses
- L7:1566 weakly suppresses
- L6:2474 weakly suppresses
- L8:268 weakly suppresses first person 'I' or 'my' in speech
- L8:805 weakly excites
- L7:278 weakly suppresses
- L7:1247 weakly suppresses
- L6:1097 weakly excites
- L6:2777 weakly suppresses
- L6:18 weakly excites
Feeds later layers
- L18:2009 suppresses
- L11:1720 suppresses
- L10:503 suppresses
- L11:895 suppresses
- L10:2175 excites
- L15:1586 suppresses
- L11:1498 suppresses
- L19:1401 suppresses
- L20:1591 suppresses
- L20:2938 suppresses
- L25:545 suppresses
- L20:989 weakly suppresses
- L10:2818 weakly suppresses
- L21:484 weakly suppresses
- L12:1889 weakly suppresses
- L16:842 weakly suppresses
- L11:288 weakly excites
- L20:2225 weakly excites
- L17:1139 weakly suppresses
- L10:1536 weakly excites
- L23:803 weakly suppresses
- L20:1150 weakly suppresses
- L21:2812 weakly suppresses
- L19:1943 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
- 导游
- Falsy
- 酩
- ottage
- ibilidade
- dựng
- difficulté
- démarch
- /sdk
- .invokeLater
Suppresses
- IRM
- cere
- Fact
- brain
- 研究表明
- horm
- researchers
- scientists
- act
- optimal