Hacker News Reader

Comment by orbital-decay | Hacker News Reader

If the rumors are correct and ideological disagreement was at the core of this, OpenAI is not going to be open anyway, as Sutskever wants more safety, which implies being as closed as possible. Whether it's "public good" is in the eye of the beholder, as there are multiple mutually incompatible concerns about AI safety, all of which have merit. The future balance between those will be determined by unpredictable events, as always.

View on HN

> Whether it's "public good" is in the eye of the beholder

That's too easy an answer, used to dismiss difficult questions and embrace amorality. There is public good, sometimes easy to define and sometimes hard. If ChatGPT is used to cure cancer, that would be a public good. If it's used to create a new disease that millions, that's obviously bad. Obviously, some questions are harder than that, but it doesn't excuse us from answering them and getting it right.

The issue with giving everyone open access to uncontrolled everything is obvious, it does have merit indeed. The terrible example of unrestricted social media as "information superconductor" is alive and breathing, supposedly it led to at least one actual physical genocide within the last decade. The question that is less obvious to some is: do these safety concerns ultimately lead us into the future controlled by a few, who will then inevitably exploit everyone to a much worse effect? That it's already more or less the status quo is not an excuse; it needs to be discussed and not dismissed blindly.

It's a very political question, and HN somewhat despises politics. But OpenAI is not an apolitical company either, they are ideologically driven and have the AGI (defined as "capable of replacing humans in economically important jobs) as their stated target. Your distant ancestors (assuming they were from Europe) were able to escape the totalitarianism and feudalism, starting from the Middle Ages, when the margins were mile-wide compared to what we have now. AI controlled by a few is way more efficient and optimized; will you even have a chance before your entire way of thinking is turned to the desired direction?

I'm from a country that lives in your possible future (Russia), I've seen a remarkably similar process from the inside, so this question seems very natural to me.

A company named openai choosing safety is bad because safety means as closed as possible, is a very questionable linguistic contortion. Most unsafe things happen behind closed doors.

Ilya himself said he's against fully open source models because they're not safe enough. He's definitely against open source, my hunch is that we will see OpenAI being less open after his takeover.

Full interview here ("No Priors Ep. 39 | With OpenAI Co-Founder & Chief Scientist Ilya Sutskever" from 2 weeks ago): https://www.youtube.com/watch?v=Ft0gTO2K85A

Reply on news.ycombinator.com