first piece of a name starting 'Be'
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. 24 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 start of a name, most often a surname, place or town that begins with the letters 'Be' or 'Op' or 'Ne', such as Beaufort, Bearden or Beagle. Turning it up did not produce any clear theme in the 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.
-
on Coastal Hazards.⏎The detailed, 43-page document draws on tidal gauge measurements at five points along the eastern coast—at Duck, Oregon Inlet, Beaufort, Wilmington and Southport—as well as projections from the Intergovernment
-
specifically, the factors that may limit uptake–in order to shed light on how to improve the promotion of food literacy skills. We will also hear from Erin Beagle, Executive Director of Roots to Harvest, who will share her insights and experiences
-
wrote lives of Daniel Webster and James Buchanan.<|endoftext|>By: H Bernard Bechtel (Author)⏎206 pages, colour & b/w photos,
-
. (Image via Zach Montellaro on Flickr, CC BY-NC-SA 2.0)⏎The two Supreme Court cases, Jones v. City of Opelika 316 U.S. 584 (19
-
had this global sense that mitochondrial dysfunction may be particularly relevant to synaptic defects, but I think this [study] really provides more specificity to that story,” says Carrie Bearden, professor of psychiatry, biobehavioral sciences and psychology at
-
Origin of the Intentional Fallacy⏎The idea of the intentional fallacy was introduced by W.K. Wimsatt, Jr., and Monroe C. Beardsley in their book
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: ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' '
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
about seven inches, tail a little more; and the Brown or Norway Rat (Mus decumanus) - the "Hanoverian "Rat of Waterton -
-
definition, we find that a pattern of distressing actions or demands<|endoftext|>Family Name: Passifloraceae⏎Botanical Name(s): Passiflora Incarnata⏎Popular Name(s): Maypops, Wild Passionflower and Apr
-
37th UGM doctoral graduate.⏎In conclusion of her dissertation entitled "Biosystematics of Murricata L. Annona, Annona squamosa and Annona Reticulata by Numerical Approach", it
-
a genus from the tropics, this plant has a kinship with Soursop (Annona muricata) and sugar apple (Annona squamosa). The custard-apple has become increasingly scarce because not many
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
- L23:635 suppresses
- L21:1771 weakly suppresses
- L19:408 weakly excites
- L0:2723 weakly excites
- L17:2018 weakly suppresses
- L9:358 weakly suppresses
- L1:1662 weakly excites
- L4:2370 weakly suppresses
- L6:49 weakly suppresses
- L19:1409 weakly excites
- L15:222 weakly suppresses
- L23:293 weakly suppresses
- L19:1946 weakly suppresses
- L21:814 weakly suppresses
- L18:1879 weakly excites
- L14:1340 weakly excites
- L12:612 weakly excites
- L4:426 weakly excites
- L19:618 weakly excites
- L20:579 weakly excites
- L12:866 weakly suppresses
- L0:1935 weakly suppresses
- L22:1668 weakly excites
- L20:704 weakly excites
Feeds later layers
- L25:1641 suppresses
- L27:661 suppresses
- L27:315 suppresses
- L27:404 suppresses
- L27:1793 weakly excites
- L25:1922 weakly excites
- L25:2646 weakly excites 'there' after a because or if clause
- L25:2121 weakly excites
- L25:2166 weakly excites
- L25:776 weakly suppresses
- L25:1855 weakly suppresses
- L27:1786 weakly suppresses
- L25:2979 weakly suppresses
- L25:383 weakly suppresses
- L27:2932 weakly suppresses capital F starting a word
- L26:1175 weakly excites
- L25:1581 weakly excites
- L27:2731 weakly suppresses
- L26:2194 weakly excites
- L25:3068 weakly excites
- L27:1457 weakly excites
- L25:3013 weakly suppresses
- L27:618 weakly excites
- L26:302 weakly suppresses colon or comma before a quote
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
- ạn
- 孑
- нес
- 赘
- encion
- airl
- יתי
- <dd
- ->__
- обраща
Suppresses
- '
- "
- ‘
- “
- ('
- 「
- «
- \"
- ("
- '[