Ask HN: Will ChatGPT have a monopoly on truth worse than search engines
While most concerns i read are about the future of jobs, any short term concerns about a monopoly over the selection of training datasets.
Does anyone have a source of a decent discussion on this topic? ChatGPT is unlikely to be the only player in the category for very long. Megacorps are all over this. MAANGs are spending $10 B's to be in this space long-term. Algorithms of all kinds do, and will, have influence over individuals to motivate them to discover, believe, do, and learn particular memes while not others. The problem then becomes an irreducible one of "teaching" an algorithms morality and ethics, critical thinking, and due-diligence. These maybe possible eventually, but are much more difficult than individual snippets of generative media. We're not at even a good point of capturing and giving algorithms specific constructive feedback from users on what's wrong with a particular sentence or image. Absolutely, this isn't limited to ChatGPT at all. Most of this space will eventually be dominated by a few players, and the niches will be filled by echo chambers of truth perhaps, similar to how modern social media space formed. Giving complete power over choice of dataset to a cooperation without any transparency over selection process is not a great idea. I sincerely hope not. I got past the safeguards early on and requested it to give me instructions on how to make plastic explosives, to do this I had to use a very complex and specific prompt. When I had the lab procedure generated, I sent it over to a friend who has a PhD in Chemical Engineering and he laughed at the instructions and said you'd likely kill yourself or cause other permeant harm to your lab and body if you attempted the instructions. He had a similar response when I asked it to give me steps for synthesizing other illicit but less dangerous organic compounds. I'm reminded of the need to supply AI with certified, authoritative information. Piss poor algorithms operating on crappy data have always produced bad outcomes. One would have thought we would have at least learned that by now. --- Garbage In Garbage Out : https://en.wikipedia.org/wiki/Garbage_in,_garbage_out a.k.a. Shit In. Shit Out. : <https://www.urbandictionary.com/define.php?term=Shit In. Shi...> Yes, yes they will. They appear to demonstrate authority. They do not merely open the door to a range of information which can face recognizable human scrutiny, traction for critical thought. ChatGPT minimizes the “attack surface” for human evaluation. This is not something we should optimize. It is unknown and potentially dangerous. Dissenting information is vital to the health of a society. Humans were not meant to drink from a single fountain of “truth,” their herd tendencies and response to social proof vastly outweigh rigor in decisions.
Well this should make it a lot easier for the government to control what the "truth" is then. > I'm reminded of the need to supply AI with certified, authoritative information.