Comment by ablyveiled | Hacker News Reader

Comment by ablyveiled | Hacker News Reader

While I can appreciate reasoning about AGI as a purely abstract thing (I loved Nick Bostrom's book, Superintelligence), it seems out-of-touch to believe it's at all likely to happen. Do these people even use GPT-4? Do they connect the dots of its capital to the dots of its performance, and see how we're already hitting a huge wall? All while losing MSFT cash and routinely failing to do even basic tasks?

I can't help but feel like I'm being marketed to. I've grown to completely distrust anything in this vein, since the reality of the system I see before me is so drastically inferior to how people appear to be reasoning about it. It's not like how a Model-T car differs from a modern one in safety, power steering, and so on -- it feels like an error of categories.

> Do these people even use GPT-4? Do they connect the dots of its capital to the dots of its performance, and see how we're already hitting a huge wall? All while losing MSFT cash and routinely failing to do even basic tasks?

Honestly it's a complete skill issue if you can't get it to "basic tasks". People are out there literally spinning up entire websites and apps in minutes using GPT. There is also no indication at all that we're "hitting a wall". I suppose next you're going to say its glorified auto-complete or a "stochastic parrot"?

Like is this guy completely blind to progress? GPT4 is no better than GPT2? That GPT5 or 6 will have even better capabilities??

And I say in turn, it's a skill issue if your "website" or "app" could be made by a GPT in a few minutes. Is what you're doing actually valuable, then? It consistently fails to solve my problems.

Yeah because only complex websites and apps deliver any value lol. I think you're really just admitting to having very poor creativity. I use LLM's everyday in systems that replace manual business processes. They're delivering clear value in a Multi-Billion dollar company. So again, skill issue.

You've got to extrapolate a bit. I mean 20 years ago AI was fairly rubbish. Now it can beat humans at all board games, and things like GTP4 can beat most humans at things like law exams. At the current rate of progress it'll be quite good in a decade or two.

Reply on news.ycombinator.com