Valuetainment - March 27, 2025


“A Threat To OpenAI” – OpenAI Whistleblower’s Parents Claim COVER-UP To Silence AI Dissent


Episode Stats


Length

7 minutes

Words per minute

172.55664

Word count

1,333

Sentence count

92

Harmful content

Hate speech

1

sentences flagged


Summary

Summaries generated with gmurro/bart-large-finetuned-filtered-spotify-podcast-summ .

In this episode, we discuss the story of a former AI researcher who left OpenAI and went to the New York Times to complain about the company's copyright practices. We also talk about how the company responded to his concerns and why they are suing him.

Transcript

Transcript generated with Whisper (turbo).
Hate speech classifications generated with facebook/roberta-hate-speech-dynabench-r4-target .
00:00:00.080 So at what point did he, before being a whistleblower, and typically as a whistleblower, to do that at a young age, that could be a career, it could hurt you with your career if you're going to come in and talk about it, right?
00:00:11.440 So you have to be careful because all the other engineering firms to hire you, they may be like, whoa, do we want to hire a whistleblower?
00:00:17.960 It's a little bit of a risk for him, right?
00:00:19.720 What got him to get to the point of saying, I have to blow the whistle here with what's going on?
00:00:24.760 Yeah, he had his own startup plans before he quit OpenAI.
00:00:30.360 In July, he had a plan to start a startup with respect to neuroscience and machine learning.
00:00:38.820 He did check with his friend from UC Berkeley who studied together, will you join me?
00:00:44.820 He already knew he was not going to be depending on anyone's job.
00:00:49.360 He's going to create his own job and his company.
00:00:51.760 And telling the ethics would demonstrate his integrity that would only help him secure funds, he said.
00:01:00.000 So this is July of 2024.
00:01:02.720 Yes.
00:01:03.140 He had already plans of starting his own company.
00:01:05.760 Yes.
00:01:06.040 He is still with OpenAI because I don't think he left OpenAI for another month or two, right after that.
00:01:11.600 Yes.
00:01:12.240 What was the exact month?
00:01:13.600 Did he leave in September or August?
00:01:15.520 August 4th.
00:01:16.420 He left August 4th.
00:01:17.640 So, and then when did he blow the whistle?
00:01:21.340 When was it when he blew the whistle?
00:01:22.700 In July itself, while he was there, we don't know the activities in the company.
00:01:27.860 But what we know is he reached out to New York Times.
00:01:30.840 So my assumption here is if he did not, if he resorted to reach out to New York Times,
00:01:38.100 maybe his voice was not heard in the company.
00:01:41.120 So he wanted to go out.
00:01:43.280 And he didn't speak to them completely, but he did reach out in the end of July.
00:01:48.960 His interview was sometime in the end of August, published October 2023.
00:01:53.940 This is the one, right?
00:01:55.440 And what's the date on this, Rob, if you can?
00:01:57.380 October 23rd.
00:01:58.320 Yep.
00:01:58.620 Right there.
00:01:59.080 October 23rd.
00:01:59.560 Okay, so the title says, can you go up a little bit?
00:02:02.520 OpenAI researcher, former OpenAI researcher says the company broke copyright law.
00:02:07.980 And in this, I've gone through this.
00:02:10.200 We've gone through the stories.
00:02:11.560 If you don't mind sharing with the audience, what is he trying to say that they, which copyright
00:02:16.140 did they break?
00:02:17.740 Yeah, he explains extremely well, like what exactly is a copyright violation and why it's
00:02:23.720 a concern and about how these LLM models, like large, like models, right?
00:02:30.020 AI models, why they are not really good.
00:02:32.920 One of the major factors is the copyright law.
00:02:36.720 That's what all these Hollywood actors, 400 of them have signed against it, right?
00:02:41.460 What they do is they have the AI engine, right?
00:02:45.260 They feed the training data to that engine.
00:02:48.860 They put all the data for it to understand and learn.
00:02:52.260 And that data comes from everywhere.
00:02:56.380 It could be journal.
00:02:57.320 It could be YouTube video.
00:02:58.520 It could be any journals, any artist's work.
00:03:01.900 When Google puts that in the page, they don't remove the copyright stuff.
00:03:06.760 But what OpenAI does is it doesn't respect the copyright law.
00:03:11.540 And also the most important declaration he makes is it changes the output.
00:03:17.060 In his research paper, he also publishes mathematically why he believes it's a change of output.
00:03:24.740 That means in a way, they're changing the work of the artist.
00:03:30.100 And that's against the copyright law.
00:03:33.580 And if anyone reads this article, he explains very clearly why it's a violation.
00:03:38.980 What is this, Rob, here?
00:03:42.020 This is, I believe, the WebGPT paper that you had discussed earlier that he wrote while in college.
00:03:49.140 Right.
00:03:49.620 Can you go back to the New York Times one?
00:03:51.520 So in the New York Times one, he's saying that over the past two years, a number of individuals' business have sued various AI companies, including OpenAI, arguing.
00:03:58.720 That's not him.
00:03:59.140 That's the New York Times saying that they illegally use copyright material to train their technologies.
00:04:05.040 Those who have filed suits include computer programmers, artists, record labels, book authors, and news organizations.
00:04:10.940 So New York Times also sued OpenAI in December.
00:04:14.400 And if I'm not mistaken, in your interview with Tucker, you were saying, you were talking about the fact that a part of the New York Times document was sealed, not open to the public, but you were waiting until the end of January for your husband to become the estate administrator for you to have access to it.
00:04:32.880 Did they end up giving you access to?
00:04:35.520 Yes, we have the access, but OpenAI did not share that information.
00:04:39.660 They say, like, there's no such email.
00:04:41.340 And the fact here is, New York Times labeled custodian witness, 10 of them on November 18th.
00:04:50.560 And OpenAI denied every one of them.
00:04:53.820 On 25th November, they said, only Suchir Balaji can be the custodian witness because he was not there at that moment.
00:05:02.220 That is strange.
00:05:03.860 That's very strange.
00:05:05.360 That's very strange.
00:05:06.200 And then also, the New York Times, at the time, if I'm not mistaken, when I'm looking at the date, November 18th, New York Times court files, filing, identified that New York Times attorneys, copyright cases on OpenAI, that certain people that would testify, and one of the names, November 18th, was Suchir that would testify, Balaji that would testify against OpenAI.
00:05:33.980 That's November 18th, okay?
00:05:37.080 And then this is, is this when he's leaving to Catalina Island for backpacking?
00:05:40.760 Is he gone already or he's about to leave?
00:05:43.020 He was already there.
00:05:44.100 He was already there.
00:05:45.420 Okay.
00:05:45.720 So, you already know that, you know, OpenAI is going through this.
00:05:50.440 New York Times is counting on the fact that Balaji is going to be speaking.
00:05:54.480 Is that the one?
00:05:55.420 Yeah.
00:05:55.660 Custodian requests to court.
00:05:57.000 New York Times argued that Balaji would have unique and relevant documents supporting to new plaintiffs allegations of willful infringement, among other issues.
00:06:05.180 The Times attorney had previously asked the court to designate eight other current or former OpenAI employees as custodians in the case, including Ilya, Suitskever, co-founder and former chief scientist, and that's the friend that you're talking about.
00:06:19.200 Yes.
00:06:19.660 And Bright Light Cap, the chief operating officer.
00:06:23.200 Nowadays, more than ever, the brand you wear reflects and represents who you are.
00:06:28.380 So, for us, if you wear a Future Looks Bright hat or a Valuetainment gear, you're telling the world, I'm optimistic, I'm excited about what's going to be happening, but you're a free thinker, you question things, you like debate.
00:06:40.500 And, by the way, last year, 120,000 people got a piece of Future Looks Bright gear with Valuetainment.
00:06:47.880 We have so many new things.
00:06:49.740 The cufflinks are here.
00:06:51.260 New Future Looks Bright.
00:06:52.280 This is my favorite, the green one.
00:06:54.140 Just yesterday, somebody placed an order for a hundred of these.
00:06:58.120 If you watch the PBD podcast, you got a bunch to choose from, white ones, black ones. 1.00
00:07:03.500 If you smoke cigars and you come to our cigar lounge, we have this high-quality, lighter cutter and a holder for the cigars.
00:07:12.040 We got sweaters with the Valuetainment logo on it.
00:07:14.720 We got mugs.
00:07:15.540 We got a bunch of different things.
00:07:16.800 But if you believe the future looks bright, if you follow our content and what we represent with Valuetainment, with PBD podcast, go to vtmerch.com.
00:07:27.460 And, by the way, if you order right now, there's going to be a special VT gift insight just for you.
00:07:32.020 So, again, go to vtmerch.com.
00:07:33.780 Place your order.
00:07:35.100 Tell the world that you believe the future looks bright.
00:07:37.820 If you enjoyed this video, you want to watch more videos like this, click here.
00:07:40.560 And if you want to watch the entire podcast, click here.
00:07:43.340 We'll see you next time.