Hacker News Reader

Comment by humansareok1 | Hacker News Reader

Given what Sam has done by clearing out every single person who went against him in the initial coup and completely gutting every safety related team the entire world should be on notice. If you believe what Sam Altman himself and many other researchers are saying, that AGI and ASI may well be within reach inside this decade, then every possible alarm bell should be blaring. Sam cannot be allowed to be in control of the most important technology ever devised.

View on HN

I don't know why anyone would believe anything this guy is saying, though, especially now that we know he's going to receive a 7% stake in the now-for-profit company.

There are two main interpretations of what he's saying:

1) He sincerely believes that AGI is around the corner.

2) He sees that his research team is hitting a plateau of what is possible and is prepping for a very successful exit before the rest of the world notices the plateau.

Given his track record of honesty and the financial incentives involved, I know which interpretation I lean towards.

...or he's just Palpatine who wants shitload of money regardless of future speculations, end of story.

This is a false dichotomy. Clearly getting money and control are the main objectives here, and we're all operating over a distribution of possible outcomes.

I don't think so. If Altman is prepping for an exit (which I think he is), I'm having a very hard time imagining a world in which he also sincerely believes his company is about to achieve AGI. An exit only makes sense if OpenAI is currently at approximately its peak valuation, not if it is truly likely to be the first to AGI (which, if achieved, would give it a nearly infinite value).

What's the effective difference between exiting now and if it does achieve in your words "nearly infinite value" to him personally?

Either way he is set for life, truly being one of the most wealthy humans to have ever exist... literally.

It's interesting because one of the points Sam emphatically stresses over and over on most podcasts he's gone on in the past 4 years is how crucial it is that a single person or a single company or a collection of companies controlling ASI would be absolutely disastrous and that there needs to be public, democratic control of ASI and the policies surrounding it.

Personally I still believe he thinks that way (in contrast to what ~99% of HN believes) and that he does care deeply about potential existential (and other) risks of ASI. I would bet money/Manifoldbux that if he thought powerful AGI/ASI were anywhere near, he'd hit the brakes and initiate a massive safety overhaul.

I don't know why the promises to the safety team weren't kept (thus triggering their mass resignations), but I don't think it's something as silly as him becoming extremely power hungry or no longer believing there were risks or thinking the risks are acceptable. Perhaps he thought it wasn't the most rational and efficient use of capital at that time given current capabilities.

Or maybe he is just a gready lier? From the outside looking in how can you tell the difference?

It's possible that both things could be true. He may be a greedy liar while still being very concerned about ASI safety and wanting it to be controlled by humanity collectively (or at least the population of each country, via democratic means).

Maybe he is only a greedy liar. I don't know. I'm just stating my personal belief/speculation.

>I would bet money/Manifoldbux that if he thought powerful AGI/ASI were anywhere near, he'd hit the brakes and initiate a massive safety overhaul.

Not sure how you can believe this given all of his recent actions and the ever growing list of whistleblowers dropping out of OpenAI explicitly saying Safety is not taken seriously.

I mean just generally the ability to actually stop and reorient around working on safety seems incredibly non trivial. To say nothing of the race dynamic he has perpetuated, the other frontier companies are unlikely to do the same.

Reply on news.ycombinator.com