Hallucinations, factual errors, a decrease in public interest, and a plunge in investment —all of those and extra have been talked about just lately in information and feedback that appear to be the harbinger of a critical setback for Generative AI. Some have even known as for the demise of the AI “bubble,” one thing they “predicted a very long time in the past.”
On this put up, I focus on how the present hurdles of Generative AI methods may very well be (have been?) mitigated with the assistance of the nice previous symbolic reasoning.
However let’s first clarify these present hurdles.
Take, for example, one thing as primary as arithmetic. I requested yesterday to ChatGPT how a lot is the cubic root of 123,456.78, getting 49.3989 as the reply. Then I requested ChatGPT to multiply 49.3989 by itself two occasions, supposedly getting again to my 123,456.78 quantity, and –no kidding– ChatGPT got here with 123,456.78 because of this, supposedly verifying the correctness of its operation. I might have believed it, however simply to make sure, I made the operation with a calculator and multiplying 49.3989 by itself two occasions acquired 120,545.73, which is nearly 3,000 items away from the meant outcome. Which means, after all, that the primary operation (the cubic root) was fallacious within the…