from Hacker News

Ask HN: Marginally viable AI-driven ideas for 2022?

by DrNuke on 1/7/22, 11:29 AM with 3 comments

Happy New Year folks, I am with three youngsters asking me for AI-driven ideas making sense in 2022. I am not sure I can help: big corps are all over industrial applications, and the internet bubble is so crowded they have no statistical chance with recommenders and content projects. Therefore, their best shot may be looking into something that would come good in two, three or even ten years from now. To put my hat into this, I am thinking of NLP and small robotics for hospices and nursing homes: low cost, low risk, high reward, prospective jobs if they can’t make it as entrepreneurs. Any different ideas? Thanks!

  • by muzani on 1/7/22, 2:11 PM

    This sounds exactly like a solution looking for a problem. Look at something that people dump a lot of effort on and make a better product.

    I find that most of the AI solutions out there are not done by people who have domain expertise. You can't really make a good tool without being good at the work. If you want to make an AI artist, creative writer, or composer, you need sufficient skill in the topic. With most of my experience with NLP, crappy input = crappy output. If you expect AI to figure it all out, they generally get it off.

    There's also a lot of non-AI related work around it, the typical plumbing, APIs, customer service stuff, and often that part can be more work. My own experiments need a lot of procedural generation work and theory. That's also another technical barrier that big and small corps have trouble with.

    The current AI buzz reminds me of 10 years back when everyone was trying to start businesses around QR codes and AR, or 20 years ago, around Bluetooth.