Pre-1900 LLM Shows Glimpses of Intuition for Quantum Mechanics and Relativity
Sonic Intelligence
An LLM trained on pre-1900 text exhibited hints of modern physics intuition.
Explain Like I'm Five
"Imagine if you only read books from before 1900, and then someone asked you about space travel. You might say some smart things that sound like modern ideas, but you wouldn't actually invent the rocket. This AI is a bit like that with physics."
Deep Intelligence Analysis
Impact Assessment
This experiment probes the nature of intelligence and discovery in AI, questioning whether current LLMs possess genuine intuition or are primarily sophisticated pattern-matching engines. It highlights the distinction between synthesizing existing knowledge and generating truly novel, out-of-distribution insights.
Key Details
- ● An LLM was trained exclusively on text published prior to 1900.
- ● The experiment aimed to determine if the LLM could derive quantum mechanics and relativity.
- ● The model largely failed at physics tasks but showed 'glimpses of intuition.'
- ● It suggested 'light is made up of definite quantities of energy' and that 'gravity and acceleration are locally equivalent.'
- ● This experiment contrasts with modern LLMs like GPT 5.2 and Gemini Deepthink solving complex, in-distribution problems.
Optimistic Outlook
Even limited intuition from an LLM constrained to historical data suggests AI could potentially accelerate scientific discovery by identifying novel connections or challenging existing paradigms in ways humans might overlook, given sufficient architectural advancements.
Pessimistic Outlook
The experiment underscores the current limitations of LLMs in true out-of-distribution reasoning and creative hypothesis generation, suggesting that achieving AGI capable of groundbreaking scientific discovery may require more than just scaling current architectures or training data.
Get the next signal in your inbox.
One concise weekly briefing with direct source links, fast analysis, and no inbox clutter.
More reporting around this signal.
Related coverage selected to keep the thread going without dropping you into another card wall.
