the '0' in years like 2050
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. 22 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 zero in future years written as 2 then 0, such as 2050 or 2030, while the model is still reading the number. On the other side it fires negatively on 'in' just before 'the future', and turning it down made the model repeat the word 'future'.
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.
-
similar study did find that cats decl<|endoftext|>With the human population expected to grow by billions over the new few decades, the United Nations is predicting that by 2050 our food production must double
-
, it released a report titled ‘Sharing the effort under a global carbon budget’.⏎WWF says: “A strict global carbon budget between now and 2050 based on a fair distribution between rich and poor nations has
-
Climate Change, to forecast sea level changes over the next 30 years. At Duck, the hardest-hit tidal gauge, the panel projects that by 20
-
below 2 degrees Celsius.”⏎The report is based on research and shows different ways to cut global emissions by at least 80% globally, by 2050, and by 30% by 2030,
-
would have been excessive anyway.⏎According to China’s 5-year plan, their target for coal-powered capacity is 1,100 GW by 2020. But by completing all their power plant projects, they would have hit
-
, which includes measures aimed at achieving the president’s goal of net zero emissions by 2050 and a completely carbon-free domestic power industry by 2035.⏎The package may include new tax credits for investments in renewable energy generation
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: the future in the future in the future. The future in the future in the future the future in the future in the future in the future the future in the future the future in the future the future in the future in the future in the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
an ash cloud over Alaska and lost all power before pilots were able to restart the engines. Ottawa wanted to reassure passengers that steps were being taken to avoid such crises in the future. So did other countries.⏎Twenty years later, the system was working
-
the cause of miscarriage remains unknown.⏎Though one in five pregnancies end in miscarriage, having a miscarriage does not make a woman more likely to miscarry in the future.⏎Some people use the word "miscarriage" to describe a
-
measures what matters. It tells us how well nations are doing in terms of supporting their inhabitants to live good lives now, while ensuring that others can do the same in the future, i.e
-
to research from RateSetter, with 53% doing something which may actually harm their credit score – potentially making it difficult or impossible for them to borrow money in the future.⏎More than half of people
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
- L21:2675 strongly excites
- L19:553 excites
- L19:564 excites
- L16:2965 suppresses
- L18:3047 suppresses
- L21:1699 suppresses
- L18:2817 excites
- L15:2466 suppresses
- L20:2152 suppresses
- L18:2532 excites
- L14:2915 excites
- L14:2691 suppresses
- L15:2338 excites
- L20:2574 excites
- L21:794 suppresses
- L14:2516 excites
- L19:1999 excites
- L18:2178 excites
- L19:245 excites
- L7:521 suppresses
- L21:2426 suppresses
- L17:1623 excites
- L20:1878 excites
- L12:1532 excites
Feeds later layers
- L26:2717 strongly suppresses word just before 'future'
- L23:1971 strongly suppresses
- L25:2440 suppresses
- L25:2579 suppresses
- L25:1275 weakly suppresses
- L24:2353 weakly suppresses
- L25:1990 weakly suppresses
- L23:240 weakly suppresses
- L25:928 weakly suppresses
- L24:377 weakly suppresses
- L27:2277 weakly suppresses
- L26:2782 weakly suppresses
- L26:207 weakly excites
- L24:1665 weakly excites
- L23:1462 weakly excites
- L23:201 weakly excites
- L26:1578 weakly suppresses
- L26:2147 weakly suppresses
- L23:457 weakly suppresses
- L25:2512 weakly excites
- L26:2903 weakly suppresses
- L25:1968 weakly suppresses
- L23:84 weakly excites
- L23:2493 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
- @m
- @(
- @section
- itä
- @Table
- جموع
- Ridley
- inois
- @",
- 女方
Suppresses
- 未来
- future
- future
- 未来的
- Future
- 未來
- Future
- 将来
- _future
- futuro