AI doesn't meet his promises yet, Would be interesting to hear from you guys what you think about it, specially SunTzu
here a traduction from a french journalist:
Erratic Tools One reason why companies are hesitant to adopt generative AI may be that it remains highly erratic and has a high error rate. Last summer, researchers at Carnegie Mellon University’s School of Computer Science tested ten of the most popular generative AIs. They asked them to perform a series of tasks of varying difficulty, all designed to be feasible for artificial intelligence. The best performer was Claude, the AI ??from the startup Anthropic. It achieved a score of 24%. Updated versions of Gemini and ChatGPT have since scored no better than 11% and 9%, respectively. “An interesting detail is that for certain tasks, when the AI ??agent is not sure what its next task should be, it sometimes tries to be creative and take shortcuts that avoid having to do the most difficult part of the task,” the researchers discovered. This should reassure people who fear losing their jobs, the study concludes. This would also explain why generative AI is experiencing lower demand than its proponents anticipated.
Beside this, the development cost are much higher:
HSBC projects that OpenAI's annual revenue will grow from US$35 billion in 2026 to US$214 billion in 2030, but even then, its expenses over the same period will exceed its total revenue by at least US$207 billion. HSBC analysts nevertheless believe that ChatGPT will continue to be the most widely used AI in the world. Its user base could reach 44% of all adults worldwide within five years. HSBC estimates that 10% of the global population will have used ChatGPT at least once by 2025. But still: the company will not be profitable, HSBC believes. Essentially, "OpenAI will have to find US$207 billion if it wants to continue losing money," the HSBC analysts conclude. |
