Latest / Early Edition with Ryan Bridge / Oskar Howell: Technology Commentator on OpenAI scrapping its GPT-6.1 Astra model

Transcript

Oskar Howell: Technology Commentator on OpenAI scrapping its GPT-6.1 Astra model · Early Edition with Ryan Bridge

• 0:00
  Speaker 1: OpenAI is scrapping the rollout of its new model. This
• 0:02
  Speaker 1: is Astra 6.1 over safety concerns. Company says, quote, it
• 0:07
  Speaker 1: didn't quite meet the bar in terms of staying within
• 0:09
  Speaker 1: the scope of authorization. Oscar Howell is tech commentator with
• 0:13
  Speaker 1: us this morning.
• 0:14
  Speaker 2: Hey, Oscar. Hey, Ryan. Thank you for having me.
• 0:16
  Speaker 1: Great to have you. So is this like a real
• 0:19
  Speaker 1: concern or is it just because we've had so many
• 0:21
  Speaker 1: bad headlines lately that they are thinking they should do
• 0:23
  Speaker 1: something about it?
• 0:25
  Speaker 2: Well, I mean, I think it's probably a little bit
• 0:27
  Speaker 2: of both, really. Ultimately, you can't really have these rogue
• 0:31
  Speaker 2: AIs going off and doing what they want. I mean,
• 0:34
  Speaker 2: one of their recent models, I'm not sure if hacked
• 0:36
  Speaker 2: is the right word, but gained unauthorised entry into a
• 0:39
  Speaker 2: number of Australian government systems, which seems to be what
• 0:44
  Speaker 2: may have prompted the kind of rollback of this model.
• 0:47
  Speaker 2: But yeah, it seems to be a growing problem as
• 0:49
  Speaker 2: these models become more and more advanced, but definitely something
• 0:52
  Speaker 2: to be thinking about.
• 0:53
  Speaker 1: So all very well that OpenAI has come out and said,
• 0:56
  Speaker 1: we will make sure our models either don't go ahead
• 1:00
  Speaker 1: or have guardrails around them. But there's nothing stopping other
• 1:03
  Speaker 1: firms from having the exact same technology with no guardrails, right?
• 1:10
  Speaker 2: Yeah, I'd say that's a pretty good summary of it.
• 1:12
  Speaker 2: I mean, ultimately, there's very little legislation or regulation around
• 1:16
  Speaker 2: the world that kind of insulates any sort of, you know,
• 1:21
  Speaker 2: organization or government from the effect of these AIs. So, yeah, absolutely.
• 1:25
  Speaker 2: And I mean, you have seen a number of kind
• 1:28
  Speaker 2: of quite high profile security issues from both of the
• 1:31
  Speaker 2: big ones in the Western market. You've got China with
• 1:33
  Speaker 2: deep seek. Yeah, I think it's a bit of a
• 1:36
  Speaker 2: wild west at the moment.
• 1:37
  Speaker 1: Oscar, where are we at with having AI defend all
• 1:41
  Speaker 1: of the government agencies, all of the websites, everything that's vulnerable?
• 1:45
  Speaker 1: Can we not have AI do the opposite and protect
• 1:48
  Speaker 1: them all?
• 1:50
  Speaker 2: Well, isn't that the plot of Terminator? I mean, it
• 1:54
  Speaker 2: worries me a little bit, to be honest. There's an
• 1:57
  Speaker 2: old proverb by one of the managers at IBM, which said,
• 2:01
  Speaker 2: a computer can't be held accountable, so it should never
• 2:03
  Speaker 2: make a management decision. So I certainly wouldn't be putting
• 2:06
  Speaker 2: it in charge of the nukes or putting it in
• 2:08
  Speaker 2: charge of any sensitive government databases. And that's something I think, yeah,
• 2:13
  Speaker 2: like I said, I think we need to be talking about, certainly.
• 2:16
  Speaker 1: Good to have you on the show. Oscar Howell, who's
• 2:18
  Speaker 1: a tech commentator.
• 2:20
  Speaker 2: For more from Early Edition with Ryan Bridge, listen live
• 2:23
  Speaker 2: to News Talk ZB from 5 a.m. weekdays or follow
• 2:26
  Speaker 2: the podcast on iHeartRadio.