from Hacker News

Ask HN: What will the world look like when generative AI reaches equilibrium?

by echan00 on 8/8/23, 6:15 PM with 2 comments

• by themodelplumber on 8/8/23, 6:35 PM

  Yes, certainly there's an internal reward system for crafting long written emails or essays, and AI would at a guess need at least 150 years to get close to this internal-interests (energy-directed) calibrated match of subjective input + subjective output.

  LLMs currently do not do the best parts of subjectivity well at all, and the deeper these subjective systems, the worse AI gets at joining the party. So to speak. For this reason people are mainly using LLMs (whether they know it or not) to create better outlines from more basic outlines, and this outcome itself is pretty amazing really.

  However for most people there is not such a deep subjective writing-focused reward system, known or acknowledged, let alone cemented, & this holds for business purposes most of all. So depending on where you work, I think you're already seeing much of the apex of LLM-driven communications, similar to the sense in which you could arguably see the apex of internet search no later than 1998, and the global advantages of Google Search no later than 2004 or so.

  Interfacing is interesting though. I'd really like to see a kind of reverse-regex for LLMs, like let's say you're writing an email reply and want LLM prompts, and it's filled in one for you. But there's a little dialog where you can edit. You type in some code if you know it: In this case "sh*op" which means shorter, much more humorous, more open, and more personal. Or whatever the code needs to be.

  This and other interfaces or innovations within the ecosystem are the interesting side of the communications aspect for me. Eventually I think it would open a new theory of energy-propagation via communications. Not that they don't exist already, but probably new theory that's way more interesting and surprising.

  In my bias I say this after having journaled about 1K words of fiction for therapeutic-expressive stuff. ChatGPT does not know how to do this for me. There is no replacement for it. It stokes my energy and is not a chore. When I'm done I basically have a symbolic to-do list completed and more concrete ideas ready to migrate into the daily list. Watching ChatGPT do this for me is effectively meaningless. Somewhere between this and "same old business email" is a really interesting new world.

• by smt88 on 8/8/23, 6:19 PM

  I think we'll spend a lot less time communicating digitally because the unencrypted digital space will be so polluted with deepfakes and lies, it won't be usable anymore.