A recent experiment with a work version of GPT has highlighted the alarming disparity between the subsidized cost and actual compute cost of AI tasks. When tasked with a simple spreadsheet summary, the AI took 5 minutes to complete, a task that a human could accomplish in approximately 30 minutes. However, the subsidized token cost for this task was $10, with a 10x subsidy, revealing an actual compute cost of around $100.

This significant difference in costs raises serious concerns about the long-term sustainability of the current AI computing model. If the actual costs are not adequately addressed, the industry may be heading towards a major crash. The implications of such a crash could be far-reaching, affecting not only the companies involved but also the users who rely on these AI services.

Photo by Mikhail Nilov on Pexels
Photos provided by Pexels