Instruction-driven time series forecasting asks models to follow natural-language instructions while making accurate numerical predictions. A new arXiv paper, ChronoScribe, argues that this creates a fundamental tension: systems must simultaneously master language understanding and precise numerical computation.

The abstract, however, is cut off after noting that current approaches often... The full paper likely details a method for resolving this tension, but the available text does not specify the approach or any experimental results. As a result, this entry can only report the problem statement, not the proposed solution.