capital 'S' starting a name
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. 21 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 a capital S that begins a name or word split into pieces, such as the start of Sputnik, Sandoz, or Snoop. It goes quiet or negative on the letters inside older names like Shakespeare. Turning it up produced rambling text that did not clearly connect to names, so the steering adds little evidence.
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.
-
, has been unveiled in the Oeiras municipality in the Lisbon metropolitan area, the Russian embassy in Portugal told Sputnik.⏎“We will always remember the names of the first daredevils who
-
, Hollingshead came on a mission of hallucinogenic proselytisation.⏎ALBERT HOFMANN AND LSD⏎Swiss research chemist for Sandoz⏎First created LSD in 1938⏎Accidentally absorbed
-
% b. If a remainder is ever zero, it means that b⏎divides evenly into a, or a is divisible by b.⏎Try this (from Salkind, 1966): Use a for-loop and an if
-
uffs that fueled American fears about a “bomber gap” and later a “missile gap.” On October 4, 1957, Sputnik, the Soviet satellite that was the first to orbit Earth, shocked Americans
-
's not only a lodestar of hip-hop, but a global cultural phenomenon. Recent attempts have even been made to resurrect him: He performed in CGI form with Snoop Dogg and Dr. Dre in 2012 and conversed
-
about the research than about the paper. Find Scholarly Journals - Writing Your Research Papers - SJSU ... Scholarly articles are kept in databases, usually by discipline or subject
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: it's something to go with a project. A project is a short-song. So, if a project is something to go with a project, then a project is something to go with a project. So, it's a recursive self
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
THE SURVIVAL of the UNLIKE.⏎Sir M. FOSTER, K.C. B., M. D., and LEWIS E. SHOKÉ,⏎LESSONS IN ELEMENTARY PHY-⏎PHYSIOLOGY. By Sir
-
L. partire, partiri, to divide, fr. pars part.⏎1. To part; to divide; to separate. [Obs.] --Shak.⏎2. To go forth or away; to quit, leave, or separate
-
x. 30.⏎Ere thou from hence depart. --Milton.⏎He which hath no stomach to this fight,⏎Let him depart. --Shak.⏎3. To forsake; to abandon; to desist or deviate
-
where it grows as a perennial shrub. It is now grown commercially in Japan, but because of our cooler climate it’s strictly an ichinensei shok
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
- L18:81 strongly suppresses
- L20:287 strongly suppresses
- L5:204 strongly excites
- L10:233 strongly suppresses
- L19:594 strongly suppresses
- L17:515 strongly suppresses
- L17:337 strongly excites
- L2:1611 strongly suppresses
- L1:5 suppresses
- L0:162 excites
- L7:287 excites
- L2:101 excites
- L16:249 suppresses
- L17:710 excites
- L0:295 excites
- L15:89 excites
- L16:827 suppresses
- L7:300 suppresses
- L18:569 excites
- L15:114 excites
- L15:115 excites
- L4:847 suppresses
- L16:113 weakly suppresses
- L15:82 weakly suppresses
Feeds later layers
- L27:2625 excites
- L22:1434 excites
- L27:2767 excites
- L22:666 excites
- L26:2636 excites
- L27:2045 excites
- L24:696 excites
- L27:2996 excites
- L27:2050 excites
- L26:2035 excites
- L27:1918 excites
- L26:2497 excites
- L24:502 suppresses
- L27:2225 excites
- L27:1871 excites
- L26:3004 excites
- L26:1555 excites
- L26:2437 excites
- L27:2425 excites
- L26:1151 excites
- L26:2376 excites
- L27:369 suppresses
- L24:431 suppresses
- L26:1430 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
- ufficient
- atisfied
- velte
- 背后
- creenshot
- AVED
- chematic
- MOOTH
- pecially
- ensitive
Suppresses
- stations
- 纩
- FLICT
- swith
- scrim
- $s
- illed
- 沙
- tsky
- tes