gravity word pieces (gravit-)
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. 2 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 reacts to the middle of words built on 'gravit' or 'grav', such as 'Gravitational waves', 'gravitationally bound' and 'gravity' as split into pieces. It is also weakly tied to 'gravis' in myasthenia gravis. When it was turned up, the model's text did not clearly move toward physics or gravity.
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.
-
their existence a century ago, physicists have been on the hunt for gravitational waves, ripples in the fabric of spacetime. That hunt is now over. Gravitational waves exist, and we’ve found them.⏎That’s according to researchers at
-
etime ripples started to get serious.⏎But the first generation LIGO experiment, which ran for eight years, wasn’t sensitive enough. Which is understandable. Gravitational waves are minuscule — the atomic jitters that pass through our world
-
about three times the mass of the sun was converted to energy in a fraction of a second.⏎The discovery has been accepted for publication in Physical Review Letters.⏎Gravitational waves are ripples in the universe caused by some of the most energetic cosmic
-
by Gizmodo have been flying for a month, the LIGO team wanted to be absolutely certain before making an official announcement.⏎That announcement has just come. Gravitational waves were observed on 14 September 2015, at
-
for advanced students and researchers.⏎Catalogue of Galactic Supernova Remnants (SNRs)⏎Listing of over 200 supernova remnants.⏎A Gravitationally Powered Oscillatory Pulsar Model⏎Proposed advanced theory of mainstream
-
which the Milky Way and Andromeda are the dominant members. The rest of the galaxies are mostly small satellites known as "dwarf galaxies" that are gravitationally bound to these two galaxies. The Large and Small Magellanic Clouds
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: "I am a senior executive in the company, and I have been asked to join the executive of the European Commission of Consumer Reports, and I have been told to the European Commission of Consumer Reports, and I have been told to the European
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
architecture and evolved from earlier Romanesque buildings. Gothic architecture combines elements of art and buildings, theology and Gothic literature. Gothic architecture has become typified by cathedrals, fortresses, palaces and churches.⏎thai interior Gothic architecture
-
ched structure from the Gothic church is characteristic of this specific style of architecture. Gothic cathedrals were typically designed with an individual wall, which has been both decorated or
-
shrines, around which cities grew. Ancient cities usually had large temple precincts with monumental religious buildings. Many medieval cities were built near monasteries and cathedrals
-
, Early English, Early English north and south transepts, Eboracum, England, Europe, Freemasonry, glass conservation, Gothic, Gothic cathedrals, Great Britain, Great East Window, Heritage Lottery Fund, hugely significant
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 suppresses
- L0:1301 suppresses
- L0:178 suppresses
- L1:5 weakly excites
- L1:340 weakly excites
- L0:1388 weakly excites
- L1:3038 weakly excites
- L1:1910 weakly excites
- L1:178 weakly excites
- L1:2891 weakly excites
- L1:1545 weakly suppresses
- L1:1857 weakly suppresses
- L0:1484 weakly excites
- L0:2601 weakly suppresses
- L1:2318 weakly suppresses
- L1:2381 weakly suppresses
- L0:2206 weakly suppresses
- L1:486 weakly suppresses
- L0:1699 weakly excites
- L1:1110 weakly suppresses
- L1:1973 weakly suppresses
- L1:2768 weakly suppresses
- L1:905 weakly excites
- L0:949 weakly excites
Feeds later layers
- L4:743 excites
- L3:2925 weakly suppresses
- L3:1528 weakly suppresses
- L20:1656 weakly suppresses
- L20:2034 weakly suppresses
- L8:2186 weakly excites
- L7:1307 weakly suppresses
- L24:2707 weakly suppresses
- L21:1933 weakly suppresses
- L3:2223 weakly suppresses
- L12:3040 weakly excites
- L3:1318 weakly suppresses
- L16:2712 weakly excites
- L13:1033 weakly suppresses
- L19:1466 weakly excites
- L6:1952 weakly excites
- L8:2850 weakly excites
- L9:1182 weakly excites
- L4:2388 weakly suppresses
- L3:2663 weakly excites
- L16:2889 weakly suppresses
- L22:591 weakly suppresses
- L9:2450 weakly suppresses
- L13:2380 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
- 提款
- atas
- Specification
- 擒
- 遗憾
- stopping
- ))[
- ちょう
- aho
- ])[
Suppresses
- erville
- Nacht
- بيض
- gree
- paque
- irteen
- izzes
- ertain
- izzle
- ряд