Matt Levine is always interesting, but his blog on the OpenAI debacle is particularly good: archive.is/drJgZ It's hard to read this as anything but an existential disaster for the whole Effective Altruism/LessWrong/Longtermist crowd. - ThreadSky

djangowexler.bsky.social • 489 days ago

Matt Levine is always interesting, but his blog on the OpenAI debacle is particularly good: https://archive.is/drJgZ

It's hard to read this as anything but an existential disaster for the whole Effective Altruism/LessWrong/Longtermist crowd.

Comments

mrthewalrus.bsky.social•489 days ago

On the one hand, those weirdos spent far too much time close to levers of power, so good. On the other hand, they were at least trying to do what they thought would benefit humanity, which is manifestly not true of the techbro capitalists who will replace them holding those levers.

djangowexler.bsky.social•489 days ago

This does not comfort me, because their idea of "the benefit of humanity" is really far removed from the welfare of any current human. Lots of ideologies have the good of humanity as a theoretical premise.

mrthewalrus.bsky.social•489 days ago

Oh, sure. My point is that their fall will not make things better, because the 'other side' in this contest is the people who just want to use their huge piles of money to make even huger piles of money at the expense of everyone else, and are somewhat competent at it.

djangowexler.bsky.social•489 days ago

Oh, definitely. At best maybe some of their more obviously insane projects don't get funded.

djangowexler.bsky.social•489 days ago

I guess at BEST maybe we can have a conversation about the actual problems with "AI" without the oxygen getting sucked up by sci fi doomsdays.

mrthewalrus.bsky.social•489 days ago

True. Still difficult to discuss, though - even if we no longer have to hear from the folks who sincerely think they're building the Machine God and want us to stop them, there's still an awful lot of marketing and general sloshing hype around the topic.

ranamar.bsky.social•489 days ago

Apparently, I have convinced the recaptcha at https://archive.is that I am a robot, because it just keeps refreshing.

djangowexler.bsky.social•489 days ago

So the sequence here is something like:

1. General AI is going to be developed soon and poses a huge risk to humanity.*

2. Us and our rich friends are very worried about this, so we set up an AI research foundation.

3. This is ok because we, the good guys, will retain control!

djangowexler.bsky.social•489 days ago

4. We need more money, so we get some investors, but we're still in charge, so we can pump the brakes if it starts to go wrong.

5. Humanity saved!

Somehow they actually got this plan going! But...

djangowexler.bsky.social•489 days ago

The *instant* they started tapping the brakes, the big investors just seized control for themselves, legal safeties or no.

djangowexler.bsky.social•489 days ago

They can do this because technology, especially IT, doesn't work like "science" in comic books. If you develop an AI, you don't end up with *one* AI in some hidden basement you can sit on, you end up with *the capability to create them*.

djangowexler.bsky.social•489 days ago

LLMs are actually MORE restricted then most because training is so expensive, but that just limits the pool of potential creators to people with billions of dollars.

djangowexler.bsky.social•489 days ago

Once again i feel like we've been badly served by SF, since the bad AIs are nearly always *singular*. SkyNet is ONE THING, created in a comic-book accident, that can be individually stopped by the heroes these guys cast themselves as.

Comments

Posting Rules

Reply