Hacker News new | ask | show | jobs
How will personalized AI be possible in the current system?
1 points by eerop 809 days ago
Hi, how do you guys think personalized AI will be made possible, given that companies are currently independently collecting user data without a central point for it? A good personalized AI would know a lot about the user's interests, preferences, etc. For example, if someone engages with Star Wars-content on Reddit and they like to use Midjourney on Discord, then a personalized AI might suggest the user to create Star Wars style images.

What do you guys think? Any articles, podcasts, etc. on this?
2 comments
ekimehtor 809 days ago
Sadly the way things are currently going I doubt this will ever happen. User data has already been commoditized and as such most companies that are able to acquire it don't want to share it, and or cannot share it due to EULAs and the like. Possibly through mergers and acquisitions such a situation could arise but that has its own set of problems given the antitrust lawsuits we have been seeing lately, not to mention the cooling effect High interest rates have had on such transactions. I suppose a third-party tracker that users willingly install onto their computers might be able to offer up such a thing but I'm not sure there's a market for that and very well may be flagged as spyware as that's basically what it would be.

But that's just my take, and I've been wrong many times before! In fact I wouldn't be surprised if TikTok and Huawei offers up exactly what you're talking about tomorrow!

link
eerop 809 days ago
Interesting, thanks for the reply.

"Sadly the way things are currently going I doubt this will ever happen. User data has already been commoditized and as such most companies that are able to acquire it don't want to share it, and or cannot share it due to EULAs and the like."

I'm ready to dive into a rabbit hole. Do you have any resources to suggest?

link
PaulHoule 809 days ago
It is an active research topic

https://arxiv.org/pdf/1909.02803.pdf

Note that the “fine tuning” and “alignment” (RLHF) stages are much shorter than the early training so technically savvy people can already customize models.

link