7 Takeaways from the Senate's AI Hearing

0:00On today's AI Breakdown, we're going over the seven takeaways from today's Senate hearing on regulating AI. Before that, on The Brief, we cover a new open-source model from OpenAI, more funding for open-source AI companies, and much, much more. The AI Breakdown is a daily podcast and video series about the most important stories and discussions in AI. If you're enjoying it, please like, subscribe, and share.

0:25Today on the AI Breakdown Brief, OpenAI is preparing an open-source LLM. Together, it raises$20 million for open-source AI. And Amazon and Samsung are both preparing chat GPT competitors. Welcome back to the AI Breakdown Brief, all the AI headline news you need in five minutes or less. A couple weeks ago, a memo leaked from inside Google. It was a researcher who wrote and argued that Google didn't have any moat and neither did OpenAI. And the argument was effectively that open source AI was out competing Google, OpenAI, and basically anyone else who was taking a closed approach to AI. One of the things that they noticed was that they might have been distracted by trying to create ever larger models.

1:05and that in fact there were advances that were being made on smaller models that were equally impressive. A couple of pieces of news are interesting in that light. The first is that according to the information, OpenAI is readying a new open source AI model. The goal would likely to be accrue some of the benefits to their ecosystem as have happened for Facebook since Facebook's LLAMA model leaked earlier this year. AI development watchdog Gary Marcus also pointed to a new paper which suggests again that these smaller training methods might actually produce pretty surprising results. Ronan Eldon tweets, Will future LLMs be based almost entirely on synthetic training data?

1:41In a new paper, we introduce Tiny Stories, a data set of short stories generated by GPT 3.5 and 4. We use it to train tiny LMs with less than 10 million parameters that produce fluent stories and exhibit reasoning. Marcus says, Fascinating new paper, which in conjunction with the leaked Google NoMote paper, is a poignant reminder that in AI, the technology could still shift dramatically overnight. There is certainly a lot of momentum around open source AI. In the weeks following the release and then full leak of Meta's Llama model, a company named Together released Red Pajama, which is an open source replication of that model entirely.

2:14Yesterday, Together announced that they had raised a$20 million seed round to continue building out open source AI infrastructure. But of course, the biggies aren't going to go down without a fight. Everyone, it seems, is trying to build a chat GPT competitor. We've got Samsung, who's trying to jointly develop a generative AI rival. as well as new competitive AI chips. And we've got Amazon, where journalists are reading the tea leaves of job listings for engineers to suggest that Amazon has plans to add a chat GPT-style interface to its online search experience. One of the big questions for this next era of computing is whether this chat or conversational interface replaces the search interfaces that we've become accustomed to over the last 20 years.

2:52Google is certainly betting that things will change given what we saw last week at I.O., and it appears that Amazon is thinking in similar ways. Finally, this morning, we get the first hearing on AI in the Senate since the rise of ChatGPT about six months ago. The witnesses include Sam Altman, the CEO of OpenAI, Christina Montgomery from IBM, and Gary Marcus, who we just mentioned, who is a sometimes critic of the way that AI is developing. I'll have a full report about what happens at that hearing on today's full AI breakdown. That's it for the AI breakdown brief. If you're enjoying this, please like, subscribe, and share with your friends.

3:22And click the notification button so you don't miss any of these reports. I'll see you back here soon for the main AI breakdown. The Senate just held its first AI hearing in the era of ChatGPT. Here are the seven most important takeaways. What's going on, guys? Welcome back to the AI Breakdown. So as I said, today I am going over the Senate's first hearing in the ChatGPT era. The hearing was held by the Senate Committee on the Judiciary, the Subcommittee on Privacy, Technology, and the Law, and it was called Oversight of AI, Rules for Artificial Intelligence. Before we get into those seven takeaways, let's do a lightning fast review of the regulatory conversation to date.

4:00About a month ago, we started to hear that the Senate was getting interested in this problem with Senator Chuck Schumer apparently getting legislation prepared, although we haven't seen much about that yet. And then, of course, a couple of weeks ago, the White House invited a number of notable CEOs from the AI space to the White House for a conversation about the risks and opportunities. Meanwhile, over in Europe, they're working on the AI Act, which is seeming like it might be some pretty strict rules and regulations when it comes to particularly open source AI. And then, of course, you have industry actors who are advocating that it should be they who create the rules, not the U.S.

4:29government. Listen, for example, to this clip of Eric Schmidt arguing that the government can't possibly understand AI. When this technology becomes more broadly available, which it will, and very quickly, the problems are going to get much worse. I would much rather have the current companies define reasonable boundaries. It shouldn't be a regulatory framework. It maybe shouldn't even be a sort of a democratic vote. It should be the expertise within the industry helping sort that out. The industry will first do that because there's no way a non-industry person can understand what is possible.

5:06It's just too new, too hard. There's not the expertise. There's no one in the government who can get it right. But the industry can roughly get it right. And then the government can put a regulatory structure around it. Now, I watched the entire three hour or so hearing so you wouldn't have to. And what I want to do is summarize it in the broadest possible sense so that you feel like you have an idea of where this hearing leaves the U.S. regulatory apparatus as it relates to AI. And the first takeaway for me is that social media is very clearly the operating framework. The senators who are thinking about the issues of AI are directly comparing it to their experience with social media.

5:42And in particular, I think the legacy of feeling far behind on social media, feeling like the government is racing to catch up, feeling like they ceded too much control to the social media platforms, those are the big starting points for them as it relates to AI. Also, I think colors the risks that they tend to see, right? We heard, for example, a disproportionate amount in this hearing about the risk of election tampering because obviously that has been top of mind as relates to concerns about social media. Now, within the broader framework of social media, there is very specifically a specter that hangs over them, which is Section 230.

6:17Section 230, for those of you who aren't familiar, is a part of the Communications Decency Act of 1996 that basically said that platforms couldn't be sued for the content that users put on them. This has created an indemnity around platforms like Facebook for all sorts of things that it seems the U.S. government wishes they now had the ability to have recourse around. In fact, one of the things that was quite notable about this was the bipartisan agreement that Section 230 was a bad thing, when many internet advocates see it as fundamental to the development of the internet in the free and open way that it's developed.

6:48It's very clear that the government hates that they limited their own power, and that doesn't seem like something they want to repeat. Now, in terms of the specifics, Sam Altman said numerous times that he didn't believe that Section 230 applied to companies like his open AI, although he wasn't sure what the framework should be instead. The third takeaway will come as a surprise to exactly no one, which is that there is a huge range of AI understanding on display among these members of the Senate. On the one end of the spectrum, you have folks who will remind you of your older relatives, typing finger by finger to understand these newfangled gadgets.

7:17But on the other end of the spectrum, you have folks like Senator Chris Coons, who asked if constitutional AI, the approach taken by Anthropic, might be a better approach to instilling ethics in AI than something like reinforcement learning from human feedback. So obviously someone over in Kuhn's office has been doing their homework. This is pretty standard for what we see in any frontier industry, that there's going to be just a big range at the beginning. And hopefully that changes over time and people form a baseline of understanding at least. Number four, I think this is a really important section.

7:44I tried to basically rank order the set of concerns that were on display here among members of the Senate committee. Number one, I think, was misinformation. And as I said, I think that this comes a little bit from the example of social media. Number two is concerns around privacy. There was, in fact, it seemed to me a big undercurrent for interest in a national privacy law. Number three, there was a lot of discussion of copyright, more frankly than I would have imagined, particularly from people like the Tennessee senator who wanted to make sure that her constituency of artists and creators was protected.

8:14Number four, again, stemming from the social media paradigm, was monopoly and concentration risk. And then the next three were mentioned, but in surprisingly less detail than you might expect. There was discussion of jobs lost. In fact, Senator Blumenthal, the subcommittee chair, said it was his biggest concern. But beyond Senator Blumenthal, there really wasn't much talk about just how likely this was to wipe out huge swaths of jobs across the economy. There was also surprisingly little discussion of adversarial harm. In other words, people using these tools for bad. Or at least it was only mentioned in passing.

8:43It wasn't something that people drilled down on and asked actual questions about. Finally, there was almost no discussion of the actual existential risks of AI. In fact, at one point, witness Gary Marcus had to drag the conversation back to that question. Senator Blumenthal had asked Sam Altman what his biggest nightmare around AI had been, but then also asked about jobs. And at the end of responding to that same question, Marcus pointed out that Sam hadn't actually answered that same question. But when we get to AGI, artificial general intelligence, maybe let's say it's 50 years, that really is going to have, I think, profound effects on labor.

9:16And there's just no way around that. And last, I don't know if I'm allowed to do this, But I will note that Sam's worst fear, I do not think, is employment. And he never told us what his worst fear actually is. And I think it's germane to find out. Thank you. I'm going to ask Mr. Altman if he cares to respond. Yeah. Look, we have tried to be very clear about the magnitude of the risks here. I think jobs and employment and what we're all going to do with our time really matters. I agree that when we get to very powerful systems, the landscape will change. I think I'm just more optimistic that we are incredibly creative and we find new things to do with better tools, and that will keep happening.

10:01My worst fears are that we cause significant... We, the field, the technology, the industry, cause significant harm to the world. I think that could happen in a lot of different ways. It's why we started the company. It's a big part of why I'm here today and why we've been here in the past and we've been able to spend some time with you. I think if this technology goes wrong, it can go quite wrong. And we want to be vocal about that. We want to work with the government to prevent that from happening. But we try to be very clear-eyed about what the downside case is and the work that we have to do to mitigate that.

10:35Now, one final note on the ex-risk part of this conversation that I will note is that towards the end of the hearing, there was a pretty sincere moment where Gary Marcus said that this was the first time that he had sat that close to Sam Altman. And it was clear to him from his body language and his energy and everything about him that the concerns that he was sharing were sincere, that he wasn't just putting on an act, and that that was something that was easier to understand and ascertain when you were in person than it was when you hear it, for example, in an interview or on TV. It felt to me like there was a moment of mutual respect and a feeling that there was progress to be made here from very different starting points as relates to these big concerns of AI.

11:12Fifth takeaway there is huge momentum towards the building of a new AI agency. This came up over and over again, and I'm not exactly sure why. It feels to me on some level, again, to be a reaction to Section 230 and not having as much control as they feel like they need over social media platforms. But I also think it's a recognition of the scope of the challenges we'll get into in a moment. Now, one notable holdout to the AI agency kind of idea was Senator Josh Hawley, who at one point argued that a more effective deterrent to problems might just be ensuring that people could sue AI companies when things go wrong.

11:44Takeaway six, and I think this is actually one of the most remarkable, is that among this committee, there were no skeptics, none of these folks who acted even a little bit skeptical of the transformative power of this technology. This is extremely rare. When you see hearings about new technology, which happen quite frequently, you always have someone say it's overhyped or it's not interesting or people are just making a big deal out of it or it's just the media. There was none of that at this hearing. The foundational starting point was an acceptance of just how powerful, how huge and disruptive this technology was likely to be.

12:23So many times these senators have heard that some new thing is internet scale change and dismissed it because they've heard it before, but not this time. There is really a sense that this is just as big a deal as people are making it out to be. I think that more than anything else is going to color the shape of the US government's reaction. By the way, for those of you who are watching the video, what you see on your screen is what came back when I plugged in the word skepticism to mid-journey. Pretty wild stuff. Finally, number seven, despite this recognition of how significant an issue AI is, there is no clear starting point.

12:55Sure, there might be some starting values or principles, things like transparency that many senators broadly share agreement on, but by and large, there are far, far more questions than answers. In fact, I think that this might be part of why there's so much focus on creating an agency, because creating an AI agency is a way to be doing something without knowing exactly what it is that you're going to do. Of course, that's probably the worst reason to create an agency in some ways, but that doesn't change the fact that that's what people are gravitating towards. The reality, of course, though, is that this technology is here and it's not waiting for them to figure it out.

13:26It was notable in the discourse how many people from both sides of the aisle were skeptical of the viability of any sort of pause, as had been advocated by experts a few months ago. Gary Marcus was careful to point out that although some genies were out of the bottle, there were still many more genies in many other bottles that we still had the ability to keep in there. But overall, that doesn't change the fact that the feeling is clearly that this technology is on the march, that it's ascendant, and that the U.S. Congress has to act soon. I guess I will close with the last subjective note. I've watched a lot of this type of hearing, a lot of them recently in the crypto domain.

13:56This was a lot less contentious and a lot more productive than most of these hearings are. I think this comes back a little bit to point six that there are no skeptics in the room, which means that they're taking their obligation to actually figure this out a little bit more seriously than they might in other contexts and circumstances. It also might reflect the fact that the AI industry is more open to regulation and cooperation with government than other industries have been in the past because they recognize just how significant and societally disruptive what their building is. Whatever the case, it stood out when people were trying to score gotcha points because it was so few and far between.

14:26That doesn't necessarily make me super optimistic that the US government is going to get it right, but at least the conversation isn't starting with some stupid political jockeying. Anyways guys, those are my seven takeaways from the AI hearing. What did you think if you watched it? Let me know in the comments if you had other reflections or you thought there were other important things to call out. And of course, if you're enjoying the AI breakdown, please like, subscribe and share it. Go listen to the podcast and subscribe to the newsletter. Until next time, guys. Peace.