2 ms·
Unless there is another jump in quality in the next GPT-5, we have plateaued. There is still time as there has been a significant jump between GTP-3.5 and GPT-4
by csomar 2y ago
Unless there is another jump in quality in the next GPT-5, we have plateaued. There is still time as there has been a significant jump between GTP-3.5 and GPT-4. I'd say we have at max one year to find out. The fact that OpenAI keeps releasing new products means they are quite pressured by market forces to demonstrate interpolation as investors are anxious about the billions being burned on compute.
So in a nutshell, it'll boil down to the next year or so.
Regarding people who say that GPT-4 increase productivity and whatnot. That's not the point, GPT-4/Claude3.5 do increase productivity and I use them daily and frequently. The question is whether that productivity increase will match the investments made. And investors have poured too much into AI.
- Lionga 2y agoThe questions is if completely new ways of using GPT/LLMs will come up. If they are just what they are now (Copilot/Summarize Text etc.) better GPT will just have very marginal gains and 99% of investments in AI is wasted. If they are somehow useful in very novel ways maybe things look better.
- csomar 2y agoWe kind of had an explosion of wrapper companies around chatGPT. It seems, by far, text -> images -> code are the only things they are capable of and in diminishing returns order. I can't really imagine them being useful to something else because the ability to reason shouldn't be scoped to one particular field. If they can reason in math, they should be able to reason in programming or medicine too.