Within the next two years, 80% of AI compute is projected to be devoted to inference and just 20% to training.

Flawed answers and the time drain of debugging generated code are the top frustrations with the technology, according to a ...