Raw LLM Responses

As a developer I have a hard time calling it AI tbh... Generative AI currently is a novelty more than anything else. As goes for all novelties: I think most scenarios for either the worst and best outcomes are vastly exaggerated. The technology itself is not even that new. By finding new applications the public has rediscovered something that has been around for quite some time. Neuronal networks have been around for decades by now. The concept itself has been around since the 1940s and tools like Tesseract have been helping with optical character recognition since the 1980s and they have found a few actually productive usecases. Speaking from experience: They never have been and probably never will be perfect, as the solution itself is limited in the accuracy it can achieve. Determining the most probable answer to a question is a clever solution, but should not be mistaken for actual intelligence. The extend of what this technology can achieve (generative AI) has been impressive - however it still is rather a first stepping stone on the way to achieve actual AI, than the thing itself. Concerning the topic of this video: I agree with the opinion, that this is rather an investment bubble, than the big technological revolution, that marketing has promised. It will find its applications and will become normalized - most likely rather as a component in software (Such as the features presented by Photoshop), than as its own independent product. The currently presented tools have no actual understanding of why and what they are doing nor what the intended outcome might be. They have no way to contextualize, understand nor reflect their solution, not even a concept of object permanence and hence depend on external validation by an actually intelligent user. This is the central factor why I fail to see how they could act as independent agents without external supervision. This video itself has shown some examples how current models rather produce something that superficially seems "about right" than is actually acceptable and ready to use. As with OCR I see use in tools, that serve more and more specialized tasks with growing accuracy than actually solving complex problems, that involve many moving parts. To all enthusiasts: We still have not achieved perfect OCR... even after almost half a century of development. Generative AI may have reallife applications in all usecases where "about right" is good enough - but anything that requires actual precision prohibits the use of such solutions. Concerning the "Ludite" criticism - opposing and criticizing innovation is productive and necessary. A technology that can neither withstand the market, law, nor opposing forces through optimization and performance alone - is damned to fail. Whether I like or dislike a hammer does not matter. Whether people will use it to forge iron does.