LLMs Predict My Coffee
2 days ago
- #LLMs
- #Physics
- #Experiments
- LLMs were tested to predict the temperature drop of boiling water in a ceramic mug over time.
- The experiment involved pouring 8 oz of boiling water into a 1.25 lb mug at 20°C ambient temperature.
- Multiple physical phenomena affect cooling: conduction, convection, evaporation, radiation, and more.
- LLMs provided equations based on exponential decay terms, with varying accuracy.
- Claude 4.6 Opus performed best but at a high cost ($0.61 in tokens).
- Experimental results showed faster initial cooling and slower later cooling than LLM predictions.
- All LLM predictions were based on one or two exponentially decaying terms.
- Some models (DeepSeek, Grok) failed to provide answers despite charging for the service.