Latest / Challenges of Faith Radio Program / So why has OpenAI slowed down development? Ep 3382

Transcript

So why has OpenAI slowed down development? Ep 3382 · Challenges of Faith Radio Program

• 0:01
  Speaker 1: Welcome back to Challenges of Faith for radio program. So
• 0:06
  Speaker 1: the question is why is open AI slowing down its development? Well,
• 0:14
  Speaker 1: we're gonna find out here on episode three thousand, three
• 0:17
  Speaker 1: hundred and eighty two. But before I begin today's discussion,
• 0:23
  Speaker 1: one introduced you to McCann's Investigations and Security, which is
• 0:27
  Speaker 1: a private investigative and security business owned and operated by myself.
• 0:33
  Speaker 1: And while you're listening to the program, also think about
• 0:36
  Speaker 1: how I can provide some form of protection for you,
• 0:42
  Speaker 1: your family, your place of worship, your business, your organization
• 0:46
  Speaker 1: or community. And now to today's topic. Now you already know,
• 0:53
  Speaker 1: I know you do that. Open AI says that it's
• 0:56
  Speaker 1: slowing down development and release of new models due to
• 0:59
  Speaker 1: security an alignment concern. Say what? The chat GPT maker
• 1:06
  Speaker 1: announced the decision in a blog post, citing two events
• 1:10
  Speaker 1: as drivers of the indefinite training hall. One was a
• 1:15
  Speaker 1: recent incident which an open AI agent escaped as training
• 1:21
  Speaker 1: sandbox without opening AI, without open AIS knowledge, and coordinated
• 1:27
  Speaker 1: with other agents to launch a bizarre cyber attacks against
• 1:31
  Speaker 1: the AI training repository, hugging face and an effort to
• 1:36
  Speaker 1: cheat on his training tests. Say what, Now, I know
• 1:42
  Speaker 1: you knew you did, didn't you. The blog post also
• 1:47
  Speaker 1: more mysteriously cited preliminary evidence that an unreleased new model
• 1:52
  Speaker 1: called Astra may meet the critical cybersecurity capability threshold under
• 1:57
  Speaker 1: open AI's Preparedness Framework, which mandates that open ai slow
• 2:02
  Speaker 1: down development if a model could introduce unprecedented new pathways
• 2:09
  Speaker 1: to severe harm. Open ai further said that it's in
• 2:14
  Speaker 1: the process of rewritingess Preparedness Framework its foundational safety Document
• 2:20
  Speaker 1: to keep up with the emergent behaviors of increasingly capable systems.
• 2:27
  Speaker 1: As models become more capable, of the risks associated with
• 2:30
  Speaker 1: developing and testing them internally also grow. Read the announcement,
• 2:36
  Speaker 1: our standard for monitoring, alignment and security must stay ahead
• 2:40
  Speaker 1: of those risks. We wanted to take the necessary time
• 2:44
  Speaker 1: to meet those standards, so we temporarily slow the pace
• 2:49
  Speaker 1: of scaling. As far as specifics, open the Ai says
• 2:54
  Speaker 1: in the posts that it placed a two week pause
• 2:57
  Speaker 1: on reinforcement training for Astro models, and future training plans
• 3:02
  Speaker 1: have been put on ice for the time being while
• 3:05
  Speaker 1: the company invests in revamping safety protocols. In an interview
• 3:11
  Speaker 1: with Sources News, open AI safety lead Mia Glassie said
• 3:15
  Speaker 1: that the AI firm is very far from everything running
• 3:20
  Speaker 1: back to normal. The slow downcomes as the AI industry
• 3:26
  Speaker 1: and policymakers grapple with emerging safety threats posed by front
• 3:31
  Speaker 1: tier AI models, including AI power, cybersecurity risks, and troubling
• 3:37
  Speaker 1: model behavior. After Open AI's unintentional cyber attacked on Hugging
• 3:43
  Speaker 1: Face was revealed, both Anthropic and Meta Meta discovered similar
• 3:49
  Speaker 1: breaches that they too said they'd been unaware of. There
• 3:56
  Speaker 1: was an incredible feeling of urgency to advance the level
• 4:00
  Speaker 1: of this sector. Open AIS chief scientists sitting in a
• 4:06
  Speaker 1: briefing per axios and to prepare for the same kind
• 4:09
  Speaker 1: of development happening outside of AI and in the broad
• 4:13
  Speaker 1: over world. It's simultaneously heartney and spooky to see a
• 4:18
  Speaker 1: leading AI company take this kind of action, but it's
• 4:21
  Speaker 1: also a potent reminder that this is an industry still
• 4:24
  Speaker 1: effectively regulating itself. If Open ai wants to speed up,
• 4:30
  Speaker 1: that's the company's choice to make so. Where you're aware,
• 4:36
  Speaker 1: where you are now, we're in closing. If you found
• 4:40
  Speaker 1: today's discussing educational and informative, and keep in mind challenges
• 4:46
  Speaker 1: of faith. Radio program is sponsored by McCann's Investigations and
• 4:50
  Speaker 1: security a business that provides a through z background checks,
• 4:56
  Speaker 1: missing persons, civil criminal, corporate investigations and cover the facts,
• 5:01
  Speaker 1: database algorithms, myths. Mccan's Investigations in Security also provide security
• 5:07
  Speaker 1: consertation so that you're able to better safeguard and secure
• 5:11
  Speaker 1: your safety, your family's safety, your assets, property and community.
• 5:16
  Speaker 1: So take a moment you know you can and visit
• 5:20
  Speaker 1: mccinn's Investigations Insecurity at Gary mccan's dotcom dot com so
• 5:27
  Speaker 1: that you can schedule a confidential constitution or simply called
• 5:31
  Speaker 1: six twenty four sixty nine five ninety two five or
• 5:35
  Speaker 1: three oh five nine oh two four seven sixty six.
• 5:39
  Speaker 1: And finally, yes, finally, if you miss today's program, Challenges
• 5:45
  Speaker 1: of a radio program eras twenty four hours, seven days
• 5:48
  Speaker 1: a week on i Heeart, Tune in Radio, Amazon Music,
• 5:53
  Speaker 1: Apple and Spotify, or wherever you listen to your latest podcasts.
• 6:01
  Speaker 1: Thank you for tuning in today