the '2' in CO2
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. 0 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 reacts to the number 2 in 'CO2', the chemical symbol for carbon dioxide, mostly in texts about climate and fuels. Turning it up did not visibly change the model's writing in a clear way.
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.
-
transfer hydrogenation chemistry and electrochemical reduction of CO2 – to develop new approaches to CO2 activation and electrochemical hydrogenation at low
-
decade in chemistry will be used to discover new classes of energy-efficient electrohydrogenation catalysts. Research will focus on three areas– chemical hydrogenation of CO2,
-
climate change, the electric power industry’s emissions of carbon dioxide, CO2, have risen 5.9 percent since 2002 and
-
and releases carbon dioxide.⏎The rate of respiration decreases with temperature.⏎Photoinhibition means the decrease in photosynthesis due to⏎exposure to excess of CO2.⏎exposure to shortage of soil moisture.⏎exposure to high temperature.⏎ex
-
to drive the chemical reaction between a molecule and hydrogen.⏎The goal of this project is to develop technologies that use electricity from renewable sources to (1) convert CO2 into carbon-neutral fuels, such as methanol, or (2) transform biomass
-
relies a lot on nuclear power: 289 g of CO2 equivalent per kWh of electricity for Germany versus just 44 g for France.
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 have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. 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.
-
Savannah, Georgia after Jacob encountered financial difficulties. It is not known how much formal schooling Phoebe had, but she was clearly well read and worldly. In 18
-
716"⏎Mercury oxide battery⏎List of books: Mercury<|endoftext|>August 18, 1823, is the birthdate of Phoebe Yates Levy Pember, the Charleston, South Carolina-born Jewish nurse and administrator
-
ati:⏎Next, the goddess created the seven planetary powers, setting a Titaness and a Titan over each. Theia and Hyperion for the Sun; Phoebe and Atlas for the Moon; Dione and Crius for the planet Mars
-
, lamp technology more or less stayed static, and no wonder. Experimenting with oil and fire tended to lead to explosions or fires. The betty or phoebe lamps used by American colonists were similar to the lamps used in biblical times
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
None: this is the first layer.
Feeds later layers
- L11:1121 weakly suppresses
- L25:1036 weakly excites
- L16:1241 weakly suppresses
- L12:1073 weakly excites
- L19:2845 weakly suppresses
- L15:221 weakly suppresses
- L4:1755 weakly excites
- L8:115 weakly excites
- L2:1239 weakly suppresses
- L18:431 weakly suppresses
- L8:222 weakly suppresses
- L10:839 weakly excites
- L14:150 weakly suppresses
- L2:401 weakly excites
- L1:1851 weakly suppresses
- L14:2786 weakly excites
- L25:1734 weakly excites
- L15:133 weakly suppresses
- L9:2176 weakly excites
- L13:2910 weakly excites
- L10:277 weakly suppresses
- L23:860 weakly excites
- L23:1914 weakly suppresses "the" after time-period prepositions
- L22:1189 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
- wre
- urga
- barg
- dem
- rece
- urge
- medicine
- revers
- zá
- ACES
Suppresses
- .fx
- 个人观点
- otch
- вин
- ipheral
- opic
- ync
- шин
- inyin
- 小心