Ask HN: Are you buying into GPT hype more than you thought you thought u would? | Hacker News Reader

Ask HN: Are you buying into GPT hype more than you thought you thought u would?

I've been in tech for 20 years. 12 years in the SF Bay Area. I'm not new to hype cycles. I've made money in some form of every main "wave". But never really believed in any of the underlying tech that much. But something feels a bit different to me about the GPT land grab going on now. I'm thinking of actually helping hype the wave myself rather than just riding a bit and watching with some ambivalence. Have I finally succumb to this place or do people really believe this is going to be big and great for the world (plus creating a mess for education/plagiarism)?

View on HN

I see major impacts it’s immediately having that warrant the hype: - as a programming tool it completely replaces google+stack exchange. Although if it kills a community like stack exchange it does raise the question where will chatgpt learn future programming questions answers? Github copilot might be a better longterm bet there. - as a writing/drafting/summarizing tool it saves a lot of time getting started writing something, or researching a topic. - It could also be really useful as a second language learning tool.

It does return invalid results quite often, but so has google for the last 20 years. And like google when it returns the right answer it’s such a time saver that it is a complete game changer. The ai accuracy can improve and there will be a lot of financial and social pressure to improve them.

I’ve been in tech over 30 years, to me it’s pretty obvious this will be as impactful as Google was, and probably more.

While I do think the hype is somewhat warranted, it's still ultimately hype. GPT is useful but not as revolutionary as people think.

Even as GPT continues to improve, mistakes will never be completely unavoidable. In fact, mistakes can be quite insidious because GPT is so often right, lulling users into a false sense of intelligence. (Especially the current iteration, which can be quite stubborn and confidently wrong even when confronted about mistakes, although this might improve in future models.)

I suspect that much like machines in the field of translation, GPT will help improve and speed up many processes, but still require human oversight.

It is different. Most, if not all, the chat apps I've used before were little more than gimmicky waste of time. ChatGPT, however, is useful. It has changed the way I work. The last app that changed the way I do my work was Google search. ChatGPT helps me with email, memos, proofreading, and idea generation. I suspect there will be more uses, but even if that's it, it offers a lot of utility. It is a big deal. My suspicion is that a lot of apps will eventually use the GPT API to enhance their functions. It's only going to get more useful as people find new uses.

Once more popular,there is a method called "programmed learning" where, based upon your answers to intermittent test questions, you might repeat a lesson or move ahead. Or, private tutors who would handle multiple subjects were once(?) more of a thing. Seems to me there is a brakthrough level of potential here, so ... yes.

Will the potential be realized, be cost-effective for the masses, etc? Not so sure.

It seems like a major breakthrough and hype is natural - it’s still early stages and I’m sort of surprised there isn’t more hype. Something like 3D printed houses seem much more hyped with less substance IMO

Been playing with chatgpt for awhile, it does a good job at making small bash/perl linux programs and turning howto's into bulletpoint lists.

I've seen it wrong on book details like characters and motivations, even when wikipedia and goodreads shows its wrong. Actually calling me out as clearly wrong until I pasted the wikipedia article. No idea if it reads the wiki articles or just said Sorry.

Its really bad in some areas, programmed to avoid details about communism like motivation, genocide and death totals. Think its avoids placing blame in its design. So history appears to be a weakness.

Asked it write stories, and it cant think outside the box much, its really limited to the input of what you tell it.

Even including stable diffusion, I think artists and writers wont have any major competition.

Document writers will be the first to be replaced. I suspect it can scan code and write documents and git comments well enough. Anything document parsing related will be the best area for instant use.

I'm excited about it's ability to translate natural language into instructions that applications understand. There are many complicated user experiences that could be simplified.

Reply on news.ycombinator.com