"discrete" and "dispersive" in technical text
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. 3 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 lights up on technical words that start with "discrete", "dispersive" or "dispersed", such as "discrete math" and "energy dispersive X-ray". It turns negative on the name "Curie" as in "Marie Curie". Turning it up did not produce a clear visible theme in the model's 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.
-
to Graphs⏎A formal introduction to graphs may be found in most discrete math textbooks. The purpose of this section is not to provide such an introduction.
-
analyses of experiments to improve the understanding of the phenomenology.⏎The Large Atomic/Molecular Massively Parallel Simulator (LAMMPS) is a Discrete Element Method (DEM) code developed at Sandia National Laboratories. The code is
-
and innovation. Analysis of the economic impact of Waxman-Markey projects that by 2035 the bill will<|endoftext|>Electron Microscopy with Energy Dispersive X-ray Spectroscopy⏎The ability for an aerosol particle to accom
-
scanning (SEM) or transmission electron microscopy (TEM) coupled with energy dispersive x-ray spectroscopy (EDX). This technique can provide not only information on
-
Students are expected to develop facility with mathematical language and symbols.⏎Because of the central and prerequisite position of discrete mathematical ideas within the field of computer science, study of discrete math is required. If a student does not have prior learning in discrete mathematics,
-
- The terms of an arithmetic sequence are said to grow “linearly.” The terms of a geometric sequence are said to grow “exponentially.”⏎- The discrete model of linear growth is ; where is the first term and the common difference.
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: to be a part of the world, to be a part of the family. So, I can be the best I can do, and I can be the best I can say, and I can be the best I can do. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
these 2 projects forms a solid basis to support policy initiatives in Flanders as well as several scientific projects (e.g. Leuven.cool or Curieuzeneuzen). The statistics derived from the garden map have already been designated
-
age of 18 living with them, 65.2% were married<|endoftext|>Wellcome to Cornucopia!⏎Cornucopia is a Marie Curie Initial Training Network that started in february 2011, funded by
-
Their Life Work Ended Their Lives⏎Their work was for the benefit of the world, but they died as a direct or indirect result of their inventions. Marie Curie won
-
Biquard, translated by Geoffrey Strachan (1966). Frédéric Joliot-Curie: the man and his theories. Eriksson. p. 129
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
- L0:1224 weakly suppresses
- L1:1504 weakly excites
- L0:178 weakly suppresses
- L0:1301 weakly suppresses
- L2:122 weakly excites
- L2:1657 weakly excites
- L0:2850 weakly excites
- L2:2081 weakly suppresses
- L2:75 weakly suppresses
- L2:70 weakly excites
- L2:3003 weakly excites
- L0:419 weakly suppresses
- L0:17 weakly suppresses
- L2:2136 weakly suppresses
- L2:2390 weakly excites
- L2:2767 weakly excites
- L0:2007 weakly suppresses 'key' after 'The'
- L0:3006 weakly excites
- L1:92 weakly excites
- L2:268 weakly excites
- L2:1673 weakly excites
- L0:2262 weakly excites
- L1:433 weakly excites
- L0:1887 weakly suppresses
Feeds later layers
- L4:3030 excites
- L4:1458 suppresses
- L15:1010 weakly excites year digits inside 20xx dates
- L5:1827 weakly excites the 'D' in 'vitamin D'
- L4:2854 weakly excites
- L14:1458 weakly suppresses
- L23:1141 weakly suppresses
- L18:1562 weakly excites
- L6:1108 weakly suppresses
- L4:2825 weakly suppresses "to" after 'for X' infinitives
- L17:558 weakly suppresses
- L11:1361 weakly suppresses
- L22:1053 weakly suppresses
- L6:520 weakly suppresses
- L12:2634 weakly suppresses
- L15:2029 weakly excites
- L5:2724 weakly suppresses
- L23:798 weakly excites
- L4:794 weakly suppresses
- L15:1407 weakly excites
- L5:2614 weakly suppresses
- L17:1870 weakly suppresses
- L18:1887 weakly suppresses
- L9:1812 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
- 流产
- addafi
- kul
- 筹码
- awl
- 好吧
- Explosion
- lier
- 要害
- ضة
Suppresses
- alem
- 命名为
- cro
- enames
- erged
- ahead
- Tul
- plement
- plemented
- 命名