Are you implying that you’re the same person I was commenting to or are you just throwing your opinion into the mix?
Regardless, we’ve seen accuracy of ~98% with simple context-based prompting across every category of generation task. Don’t take my word for it, a simple search would show the effectiveness of “n-shot” prompting. Framing it as “it _can_ reduce” hallucinations is disingenuous at best, there really is no debate about how well it works. We can disagree on whether 98% accuracy is a solution but again I’d assert that for >50% of all possible real world uses for an LLM 98% is acceptable and thus the problem can be colloquially referred to as solved.
If you’re placing the bar at 100% hallucination-free accuracy then I’ve got some bad news to tell you about the accuracy of the floating point operations we run the world on
Regardless, we’ve seen accuracy of ~98% with simple context-based prompting across every category of generation task. Don’t take my word for it, a simple search would show the effectiveness of “n-shot” prompting. Framing it as “it _can_ reduce” hallucinations is disingenuous at best, there really is no debate about how well it works. We can disagree on whether 98% accuracy is a solution but again I’d assert that for >50% of all possible real world uses for an LLM 98% is acceptable and thus the problem can be colloquially referred to as solved.
If you’re placing the bar at 100% hallucination-free accuracy then I’ve got some bad news to tell you about the accuracy of the floating point operations we run the world on