Comment by charbzg | Hacker News Reader

Comment by charbzg | Hacker News Reader

> There’s an odd feeling that whatever I am going to build will just get gobbled away by some big tech company.

I have been wrestling with this exact feeling (and everything you posted) since the first day they released chatGPT to the public. As a software engineer / Maker, I can't possibly overstate how essential it is for me to be in the flow, writing software, building things and finding novel solutions to problems. If you take this away from my day-to-day job I have no idea what I would do with myself. I know I would be absolutely miserable.

People who are blindly "enthusiastic" and expressing their excitement about being able to "build things they never built before" "with such ease" are totally missing the mark. They're just being short-sighted and can't think about the logical implication of what it means when (in a few years) Products can be created simply by telling a computer what you wish for. Here is what happens then:

(1) The value and uniqueness of all products drops to near-0

(2) You can't create a business out of these "products" because anyone else can build them

(3) The only one profiting is the person or entity that owns the AI platform you're building your products on.

Why are people not revolting against companies who's mission statement is to "build AGI" is beyond me. it's just humans being completely oblivious as always. it's not like we haven't seen this with climate change.

I think you bring up some good points.

One thing I'd add: If we enter a world where the ability to build some product, movie, song, novel, whatever is so widespread and cheap that anyone can do it (and I do emphasize the word "if", because I think its debatable that this will ever happen at a level that is cost-accessible):

The strongest way businesses will be able to differentiate is Taste. Human-ness; novelty, Design, things that AI isn't actually all that good at. AI may generate an episode of Law and Order, but it probably couldn't generate, say, Oppenheimer; because nothing in the world knew that it wanted what Oppenheimer delivered. If everything else goes to zero, then that's all that remains.

The issue is: AI is tremendously bad at working with Humans who actually have taste. You generate an AI image with the subject on the left; you ask for a revision, same image, but move the subject to the right side of the image; it changes too much, or generates new images entirely rather than iterating. Its not-obvious that this will get better because it is somewhat antithetical to the way the AI works.

Put another way, english lacks precision. Its not the right language for this. And what seems inevitable to me is that we'll need something other than english to get the last 20% finished with anything AI is being used to deliver. That thing is, guess what, a human knowing python, knowing After Effects, knowing Photoshop, etc. An image generating AI will be popular; a .psd generating AI, with extremely precise layers left in-tact, would be worth billions.

In that sense, I think the current phase of AI is absolutely "people with no taste now being able to lean on the trained taste of everyone else", but you're right that they're totally blind to one IMO non-zero probability endgame: Taste + Skill + AI becomes a hyperweapon. Creation (AI), in the right direction (Taste), with the technical acumen to take the 80% output AI generates up to 100% delivery (Skill).

The unsolved problem is: How do we integrate AI into that workflow in a way that puts the Human, that puts Taste, first? Such that it isn't "AI does 80%, then passes it off to a human", but rather that the AI becomes more like Photoshop, or After Effects, or VSCode, involved in the process from Day 0, integral to it, back and forth. No one has solved this. No one is even remotely close.

> AI is tremendously bad at working with Humans who actually have taste. You generate an AI image with the subject on the left; you ask for a revision, same image, but move the subject to the right side of the image; it changes too much, or generates new images entirely rather than iterating. Its not-obvious that this will get better because it is somewhat antithetical to the way the AI works.

You can mark parts of a generated image in Dall-E/ChatGPT and tell it how to change it. That’s generally called inpainting.

And you can iterate on the same image. For example by leaving the seed unchanged while experimenting with the prompt. Or by using image to image generation instead of text to image.

And then there are the endless ways to change the style of an image to get a desired effect.

And so on. So I personally wouldn’t subscribe to your hypothesis that these tools just show the average taste and people with their own taste can’t use them.

Looks like we found of the "people with no taste now being able to lean on the trained taste of everyone else" people I was talking about.

> (in a few years) Products can be created simply by telling a computer what you wish for

Most people don't know what they want though, and we take it for granted because we are good at making things that others are too.

We'll just get people producing a bunch of products that all feel the same (i.e. what AI is trained on).

Yes it'll get easier to build things - but aren't many things already much easier? I mean go back in time and make a website then compare it to the tools today.

I feel like we're not giving humans credit for true creativity. The bar is being moved, barriers are being changed but don't underestimate the resilience of people.

There are and will be many opportunities, we just need to use our natural intelligence to get there.

Posts like this are dangerous because they could blunt the rapid investment in AI if enough people read them. HN should do something to hide them.

You have nailed it.

Reply on news.ycombinator.com