first piece of split proper names
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. 26 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 fires on the first chunk of a name or place that the model has split into pieces, such as the "Or" in Origen or the "Merr" in Merrimack. It goes quiet, and slightly negative, on words like "it" or "but" that come after a comma or a clause. Turning it up made the text drift into odd repeats of the word "need" or "light", which does not clearly connect to names.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
) testifies (Adv. Har. xxix.1) It is derived from the Hebrew Ebion, "poor," and was not given, as Origen supposes, for their low view of Christ. Id. at 6
-
. However, right around 235, Hippolytus was reconciled with the Church.⏎One of the other items on Pontian’s plate was Origen. As a young man, Origen established a school of philosophy in Alexandria
-
Saint-Galmier (Loire).⏎Seven years later, he found a partnership with Bessac and the painter Joséphus Veyrat. Then, he set up his own business in Grigny,
-
Boulard can offer gingivitis treatment in Merrimack, NH. When left undiagnosed, gingivitis can
-
Once you develop a familiarity with topics such as the difference between regression and classification models, you’ll be able to solve an array of machine learning problems. Author Scott V. Burger provides several examples to help you build a working knowledge of machine learning.⏎
-
Phormidium pseudpriestleyi.⏎Phormidium pseudpriestleyi isolated from a meltwater pond on the McMurdo Ice Shelf, Antarctica© Patricia Mesquita⏎Microbial mat with Ph
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: light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light light
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is based on objective investigation, reasoning and analysis as opposed to subjectivity and prejudice, belief and myth. Evidence may be derived from a purely scientific investigation, but it has to be allied to the political, cultural, economic and social dimensions of these
-
. Think fresh fruits and vegetables, nuts, seeds, oils, and whole grains. While a whole foods diet does focus on eating lots of plant-based foods, it does not necessarily mean you have to become a vegetarian or vegan. Meat and dairy
-
fertilizers 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
-
quialism, exclamations, repetitions, and circumlocutions—in short, all the features of a mind instantly verbalizing. In this way, the reader experiences with freshness and immediacy the thoughtless ease with which they wound each other
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
- L16:466 suppresses 'is' after a long subject phrase
- L15:184 suppresses
- L25:586 weakly excites
- L19:2324 weakly excites
- L1:81 weakly excites
- L21:262 weakly suppresses
- L18:541 weakly suppresses
- L21:2710 weakly excites
- L21:1657 weakly excites
- L23:176 weakly excites
- L17:3018 weakly suppresses
- L21:456 weakly suppresses
- L24:241 weakly excites
- L18:233 weakly excites
- L19:545 weakly suppresses
- L4:450 weakly suppresses
- L9:74 weakly suppresses
- L22:309 weakly suppresses
- L23:287 weakly excites 'are' after a plural subject
- L11:840 weakly excites
- L24:401 weakly excites 'it' referring back to a noun
- L16:184 weakly excites
- L19:2398 weakly excites
- L19:2002 weakly suppresses 'is' or 'was' after a subject
Feeds later layers
- L27:1493 strongly suppresses
- L27:1123 suppresses
- L27:1202 weakly excites
- L27:2598 weakly excites
- L27:883 weakly suppresses
- L27:1797 weakly suppresses
- L27:2576 weakly excites capital N starting a name
- L27:1301 weakly excites
- L27:1234 weakly excites
- L27:935 weakly suppresses
- L27:1177 weakly suppresses
- L27:2527 weakly excites
- L27:2767 weakly suppresses
- L27:1030 weakly excites
- L27:2845 weakly excites
- L27:2635 weakly suppresses
- L27:1239 weakly excites
- L27:29 weakly excites
- L27:310 weakly excites
- L27:2698 weakly suppresses
- L27:3059 weakly suppresses
- L27:758 weakly excites
- L27:1771 weakly excites
- L27:1514 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
- pio
- ARDS
- เฉพาะ
- rades
- からは
- odia
- rire
- ients
- marshall
- デン
Suppresses
- need
- need
- Need
- Need
- _need
- NEED
- 需要
- needing
- 需要用
- want