What the Senate Wants to Know About OpenAI

0:00Today on the AI Daily Brief, a deeper dive into the recent letter received by OpenAI with some pointed questions from a group of US senators. Before that in the headlines, Elon asks if Tesla should invest$5 billion into XAI. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in the show notes. Welcome back to the AI Daily Brief Headlines Edition, all the AI daily news you need in around five minutes. Once again, we kick off today with a provocative tweet from Elon Musk. On Tuesday night, Elon tweeted, Should Tesla invest$5 billion into XAI, assuming the valuation is set by several credible outside investors?

0:41Board approval and shareholder vote are needed, so this is just to test the waters. With 764 ,000 people voting at the time of this recording, 68.4 % said yes, 31.6 % said no. Some of the more popular responses, Robert Scoble writes, doing so will put Tesla ahead for decades, so yes. Beth Jezos of EACC fame writes, absolutely aligns interests and will enable massive benefits from combination of multimodal data sets between internet data from X and embodied intelligence data from Tesla Optimus and the Tesla fleet. We'll make both products better. Now, what's going on in Tesla that might be driving this?

1:15The Wall Street Journal published a piece this morning, Tesla's auto woes crash Elon Musk's AI dreams. Second quarter results show a challenged car business while the CEO just wants investors to think about autonomy. Tesla, which is down 12 % at the time of recording, according to WSJ, quote, finally has financial results to support the argument that it is more than a car company. They still don't go anywhere near justifying the latest run-up in the stock. And so, of course, the question which will be debated, should Elon push this farther, is what this would actually do for Tesla. Now, of course, you have to think that part of the reason that people have been interested in XAI in the first place is the potential links between it and Elon's other portfolio of companies, most notably Tesla.

1:53And so really, I think this is a question not so much of whether the XAI folks and investors would think this is a good idea, but whether retail Tesla investors think it is a good idea or just a distraction. As we discussed yesterday, Meta has officially launched its Llama 3.1 suite, including for the first time its big flagship model, the 405B. More interestingly, however, Meta's CEO Mark Zuckerberg has been absolutely stumping for open source. As part of the announcement, he published a blog post, Open Source AI is the path forward. He writes, In the early days of high-performance computing, the major tech companies of the day each invested heavily in developing their own closed-source version of Unix.

2:29He basically argues that he believes that AI is going to develop similarly to how Linux gained popularity, and that this is better for developers, for companies, and for the world. Over in market land, Google parent company Alphabet reported its results this week and was up 14 % year-over-year on the last quarter. Unsurprisingly, AI featured bigly in Google CEO Sundar Pichai's comments about the company. Google's cloud division hit$10 billion for the first time while also achieving a billion dollars in operating profit. And Pichai said that generative AI for cloud customers already generated billions in revenues and are being used by more than 2 million developers.

3:02The Times framed this as Alphabet reports 29 % jump in profit as AI efforts begin to pay off. They focused on that same quote from Sundar. And frankly, what's interesting to me is just the Rorschach test of how different news outlets are interpreting this. For some, like the Times, it's the story of AI efforts beginning to pay off. Others are saying things in headlines like it's just not enough. For a little while, then, I think we're just going to see AI performance results as it relates to market valuations as something that is in the eye of the beholder, or in this case, the individual investor.

3:31Lastly today, earlier this week, we used the context of Harvey, the legal AI startup, to discuss whether it was too late to start an AI startup in general or a verticalized LLM in specific. While now a couple days later, Harvey has announced that it's raised a$100 million Series C at a$1.5 billion valuation. TechCrunch writes, the bulk of the new capital will be put towards collecting and curating data to build and train domain-specific AI models, while growing Harvey's headcount and expanding its paid services to new geographies. Now, of course, once again, it's all questions of interpretation.

4:01You might remember that the information had reported back in early June that Harvey had planned previously to raise$600 million at quote, at least a$2 billion valuation, in part to acquire legal research service company Vlex, but those plans had fallen through. Now, from where I'm sitting,$100 million is nothing to sneeze at, but the slightly reduced ambition could also show that even with some big backing, this is still a very difficult road. I personally am glad that companies like Harvey are getting the capital they need to go out and try and figure out how to develop AI that can handle more advanced use cases in more complete terms.

4:33And so to them, I say congrats and good luck. For now, though, that is going to do it for today's AI Daily Brief Headlines Edition. Next up, the main episode. Today's episode is brought to you by Venice. Venice is a private, uncensored generative AI app. It accesses open source models to enable text, image, and code generation without the fear of being spied on or having your data exploited. Discuss anything with Venice without concern about it being monitored, sold, or given to advertisers and governments. Venice is different because your conversations and creations are kept securely within the browser, never stored or accessible by Venice.

5:05Unlike other AI apps, Venice won't tell you what's okay to say or not. Venice won't patronize you. It simply provides direct access to machine intelligence. No topics are off-limits, No ideas are taboo. With Venice, you're in control of the AI, as you should be. Pro subscriptions are available for$49 a year or$8 per month. AI Daily Brief listeners receive a 20 % discount on Venice Pro. Visit venice.ai slash nlw and enter the discount code nlwdailybrief. That's nlwdailybrief, all one word. Today's episode is brought to you by Super Intelligent. As you guys know, Super Intelligent is a platform we are building to help everyone, individuals and teams maximize their use of AI.

5:45We help you figure out how to use AI tools, as well as what to use AI for. And this is really important. The whole goal of Superintelligent is not just to give you tutorials and lessons, but to show you how other people like you are actually getting value from AI right now. For those of you who are still out there working, learning, and grinding deep in the summer, I'm excited to share our best offer ever. If you sign up with code YEAR50 right now, you will get 50 % off the already reduced annual price. That means you'll get access to Superintelligent for a full year for less than$100. Again, that code is YEAR50 for 50 % off the already reduced annual fee.

6:24This particular code is going to expire in just a week or two. So head on over to besuper.ai and check it out. Welcome back to the AI Daily Brief. Today, we're digging a little bit deeper into that letter that Sam Altman received earlier this week from a group of U.S. senators. We're going to put that in the context of the changing political landscape and how recent shakeups to the presidential election could impact AI policy in the United States. The senators behind the letter are Brian Schatz, Ben Ray Lujan, Mark Warner, Peter Welch, and Angus King. All of them are Democrats except Angus King, who is an independent technically, but who caucuses with the Democrats.

6:59The letter basically starts with the framework that OpenAI is a really important company in a really important sector, and so therefore it is really important that everyone understands what's going on with them. Quote, safe and secure AI is widely viewed as vital to the nation's economic competitiveness and geopolitical standing in the 21st century. Moreover, OpenAI is now partnering with the U.S. government and national security and defense agencies to develop cybersecurity tools to protect our nation's critical infrastructure. National and economic security are among the most important responsibilities of the U.S.

7:27government and unsecure or otherwise vulnerable AI systems are not acceptable. Given OpenAI's position as a leading AI company, it is important that the public can trust in the safety and security of its systems. This includes the integrity of the company's governance structure and safety testing, its employment practices, its fidelity to its public promises and mission, and its cybersecurity policies. The voluntary commitments that you and other leading AI companies made with the White House last year were an important step towards building this trust. And really what this comes down to, or appears to be to me, is a reflection on all the various controversies that OpenAI has seen over the last six or seven months, and a check-in to see where OpenAI's stated commitments lie.

8:01So the net impact of this letter is that they want information by August 13th of this year, and let's look at the specific questions they're asking and try to see what they reveal about how this group of senators is thinking about OpenAI. Question one, does OpenAI plan to honor its previous public commitment to dedicate 20 % of its computing resources to research on AI safety? What they're referring to here is last year when OpenAI announced the super alignment team, part of the more tangible commitment that came with was them dedicating 20 % of the computing resources that they had up to that point secured to the super alignment team.

8:32Now, a couple of things about that. First, it was notable even then to me that the language was 20 % of the compute we've secured to date, with I imagine the anticipation that they were going to secure a lot more compute in the months and years to come. And so even at that point, 20 % might not really have been 20 % in the way that, for example, these senators are looking at it, which doesn't distinguish between when the commitment was made and now. But beyond that, some of the reports have been that part of the reason that Jan Leakey and others from the super alignment team left is that they weren't getting those resources.

9:00I'm pretty sure that the subtext of this question is accusatory and suggesting that OpenAI is not honoring this commitment. They basically say, if you are going to honor this, what are the steps that you're going to take to honor it? And if not, what percentage of resources are you dedicating to it? The second question, once again, reveals the recent controversy genesis of this letter. Question two, can you confirm that your company will not enforce permanent non-disparagement agreements for current and former employees? This has been one of the big dust-ups over the last couple months. That is, the reports that OpenAI has acted somewhat vindictively towards employees that were leaving, saying that they would surrender their equity if they didn't sign a non-disparagement agreement.

9:37This was definitely a black-eye moment from a public perception standpoint. While many people that I talked to who had interacted with big companies like this before didn't think it was all that unusual, but for those who hadn't been through a similar process, it sort of looked like a what-do-you-have-to-hide kind of moment. Question three, can you further commit to removing any other provisions from employment agreements that could be used to penalize employees who publicly raise concerns about company practices, such as the ability to prevent employees from selling their equity in private tender offer events.

10:04This is really a continuation of that second question. And for OpenAI's part, they've already committed to making these changes in their agreements. It's something that Sam Altman has discussed and apologized for publicly. But still, of course, the bigger question underneath this isn't really about employee equity. It's more what's raised in question four from the senators. Does OpenAI have procedures in place for employees to raise concerns about cybersecurity and safety? How are these concerns addressed once they are raised? Have OpenAI employees raised concerns about the company's cybersecurity practices?

10:32Question five, what security and cybersecurity protocols does OpenAI have in place or plan to put in place to prevent malicious actors or foreign adversaries from stealing an AI model, research, or intellectual property from OpenAI? This is one of the drums that former OpenAI member Leopold Aschenbrenner has been beating. He had argued at one point in a memo that the company wasn't doing enough to prevent hacks from China. And while OpenAI have said very strenuously that's not why he parted waves from the company. That's been part of the broad perception. Question six, the OpenAI supplier code of conduct requires your suppliers to implement strict non-retaliation policies and provide whistleblower channels for reporting concerns without fear of reprisal.

11:04Does OpenAI itself follow these practices? You can sense that all of this set of questions is fundamentally about whether the channel for OpenAI employees who see something that scares them is open to actually report that, or at least how difficult is OpenAI going to make it? From there, though, we move to a similar theme, but that relates to how independent actors can interact with OpenAI to assess the safety of its models. Question seven, does OpenAI allow independent experts to test and assess the safety and security of OpenAI systems pre-release? Question eight, does the company currently plan to involve independent experts on safe and responsible AI development in its safety and security testing and evaluation processes, procedures, and techniques, and in its governance structure, such as in its safety and Security Committee.

11:46This question eight is a little bit out of the blue in that it seems to be intimating that this group of senators would like to see OpenAI involve independent experts in its safety and security testing and evaluation process, as well as in its governance structure. But you can imagine that a private company might bristle at that sort of suggestion. Then we get to the one that most people are talking about online. Question nine, will OpenAI commit to making its next foundation model available to US government agencies for pre-deployment testing, review, analysis, and assessment? Where we're starting to get in the AI space is moving away from generalities to specifics.

12:17There have been vague discussions of the government having advanced access to frontier models before they're released more broadly, but they've always been in the context of some future state of AI capability. The senators are now pushing the question of whether the time is now, and whether GPT-5 is the model to begin with. The final three questions are a little bit more blasé. Question 10, what are OpenAI's post-release monitoring practices? What patterns of misuse and safety risks have your teams observed after the deployment of your most recently released LLMs? Question 11, do you plan to make retrospective impact assessments of your already deployed models?

12:48Question 12, please provide documentation on how OpenAI plans to meet its voluntary safety and security commitments to the Biden-Harris administration. Now, it's notable that this is not a specific committee. This is just a group of senators. And so whether we'll actually see a broader conversation come for this remains to be determined. What is clear, though, I think, is that it represents a bit of a ratcheting up of the seriousness with which the U.S. government is taking these potential issues. and, depending on your perspective, a bit of a noose tightening or a coiling around OpenAI, particularly when it comes to U.S.

13:17government visibility into what they're building. Of course, not everyone in the U.S. government has the same perspective on what the U.S. government's role in AI should be. Retiring Congressman Patrick McHenry, who is the current chair of the House Financial Services Committee, recently used a hearing on AI to beat the drum that the U.S. needs to remain the leader when it comes to global innovation. We cannot allow fear of the unknown, he said, to thwart the United States' role as a hub for technological innovation, Far greater than the risks associated with AI itself are the risks of allowing foreign competitors and adversaries to leave the development adoption in terms of use.

13:47Which of course has brought many to the question of what a new potential Democrat administration, after President Biden announced his withdrawal from the race, could mean when it comes to AI. The New York Times published a piece, A Kamala Harris presidency could mean more of the same on AI regulation. The article notes that Kamala Harris has actually taken a leading role on the White his efforts on AI, which has frankly been somewhat unusual as she's been fairly sidelined when it comes to policy during her time as VP. The Times points to a speech back from November where Harris said, we reject the false choice that suggests we can either protect the public or advance innovation.

14:19We can and must do both. As many point out, given her origins in California, she has more experience with big tech as well. Many in the technology industry who have been uncomfortable with the industry's shift to supporting former President Donald Trump are now trying to rally around Kamala Harris and have suggested that she seems to be more open and less antagonistic towards technology than the previous Democratic president and President Biden. Still, right now, there's no actual information. There's been no new statements. It's not even 100 % that Kamala Harris will be the Democratic nominee.

14:48But of course, should that come to pass, as it seems like it will, it will have a big impact on how AI could be shaped going forward. And it is something that we will spend more time on on the show. For now, though, that is going to do it for the AI Daily Brief. Until next time, peace. do you