logoalt Hacker News

t-3 • today at 3:00 PM • 3 replies • view on HN

I don't think anyone is suggesting that AI isn't useful - that's been proven. What is being questioned is the business model and whether it's worth the investment necessary to train huge frontier models. How does one get ROI when models are continuously improving and being surpassed and people can distill your weights with a fraction of the cost and resources? How to insulate against the threat of local open weight models as hardware improves?


Replies

lambdasquirrel • today at 3:11 PM

This is why it didn’t made sense to make OpenAI go private, instead of leaving it as a non-profit. If there’s no moat around building a dense frontier model, and others (like Google) already have a way to monetize a LLM better than you do, and yet everyone benefits from it, then it’s best left as a shared good.

➕ show 1 reply
origini • today at 3:58 PM

I will replace one person in my Team with AI. Thats one full time salary.

I'm also sure that we will forced to leverage AI a lot more in the next few years which will increase token usage even more and it will be a lot harder for us to get a headcount or consultants.

Open Chinese models have the risk of being manipulated and they are distilling Frontier.

HArdware and compute is an issue, you need partners you are allowed to use (export bans, national security) and who can provide a certain qualith/uptime and trust. After all you are sending all the private data from everyone to them.

petra • today at 3:12 PM

It's a matter of national security at this point. So they'll probably find money.