White House Secures Responsible AI Commitment from OpenAI, Meta, Google, Anthropic and Others

0:00Today on the AI Breakdown, we're looking at the White House's new commitments from Meta, OpenAI, Google, and more. Before that on The Brief, Google founder Sergey Brin is back working with the company on their new foundation model. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our Discord, and our newsletter. If you need some evidence of how intense the AI battle is getting, look no farther. Welcome back to the AI Breakdown Brief, all the AI headline news you need in five-ish minutes or less.

0:35We begin today with a Wall Street Journal report that Sergey Brin, the co-founder of Google, has returned to the company in a major way. Now, Sergey and his co-founder Larry have never been fully disengaged, even after they left their main day-to-day executive leadership roles. However, Sergey is back not just in a nominal way, but is actually spending three to four days a week in the office. His focus? You guessed it. He's back to work on AI. From the WSJ,

1:19Sergey, they say, is working on Google's forthcoming AI model, Gemini. From the journal Sources, It sounds like Sergey is involved in technical matter discussions. The one that they cite is loss curves, which is a way of measuring an AI program's performance over time. They also suggest that he has some sort of managerial role. He's been convening weekly discussions of new AI research with employees on the team. Finally, and it seems likely to me that this might be one of the biggest places that he can add value, the WSJ writes he has also intervened in personal matters such as the hiring of sought-after researchers.

1:51In other words, it's one thing when Google's trying to hire you. It's a whole different thing when Google's co-founder, Sergey Brin, is trying to hire you. Now, the journal understands that the context is intensifying competition in AI research. Indeed, later today, I'm working on an episode all about ChatGPT's new custom instructions feature, which comes just a week and a half after they launch Code Interpreter. The pace of not only development, but feature release is blistering right now. Gemini says the journal is Google's attempt to build a general-purpose AI that can rival GPT-4. The project is being overseen by Demis Hassabis, who's the CEO of DeepMind, which has recently blended with another Google group to be the singular AI body within the company.

2:29The other big detail from the piece is that there are some indications that Gemini could be released by the end of the year. You have to think between OpenAI's continued development and Meta this week announcing their commercial Llama 2, which, by the way, was announced with a big partnership with Microsoft, there has got to be a lot of internal pressure at Google right now. Next up, one of the big themes of this year has been the constraints in AI compute. It's led to a massive rally for NVIDIA, which of course dominates the AI chip production market. One of the companies that is trying to challenge them is called Cerebris, and yesterday they announced something that they call their Condor Galaxy 1.

3:06It's a 4-exaflop AI supercomputer. Now, as part of the announcement, they also share that G4-2, a UAE-based company, was ordering three of the Galaxy One systems with an option to buy six more. The deal in total was worth more than$100 million. Investor Brad Gerstner tweeted, Accelerated compute is a massively constrained resource. As shareholders of both NVIDIA and Zerebris, we celebrate this competition, which gives governments and enterprises more choice in AI compute. Moving to the policy realm for a moment. On our main episode today, we're going to talk extensively about the new commitments that the White House has secured from seven leading AI companies about how they will develop more responsibly and openly.

3:47But in the meantime, senators continue to roll out more specific AI bills as well. Pennsylvania Democrat Bob Casey introduced two new bills yesterday that both deal with artificial intelligence in the workplace. The first bill is called the No Robot Bosses Act, and basically it would prevent employers from using entirely automated systems in order to make employment decisions. So they couldn't just farm out hiring decisions to an AI which determines who's the best suited for a particular position or not. The bill would also require employers to train their employees on any systems like that that they use, and would also require human oversight of any automated AIs that are used to make employment-related decisions.

4:22The summary of the bill read, Systems and software, not humans, are increasingly making decisions on whom to interview for a job, where and when employees should work, and who gets promoted, disciplined, or even fired. Maybe you're a delivery driver and your employer's tracking algorithm determines you're not performing to its standards, and then sends you an email to let you know you've been fired without any warning or opportunity to speak to a human being. In 2023, these scenarios are no longer just imaginary. Without oversight and safeguards, these robot bosses increase the risk of discrimination, unfair disciplinary actions, and dangerous working conditions.

4:53Now, the second bill from Senator Casey has similar themes. It's called the Exploitative Workplace Surveillance and Technologies Task Force Act, and it would create a task force, a government body, whose job it was to examine and study workplace surveillance tools. The bill would, quote, analyze the effect of workplace surveillance and automated decision systems on employee salaries and scheduling, determine the impact on employee organizing efforts, and evaluate how such systems affect historically marginalized populations in the workplace. Speaking to both the bills, Senator Casey said, I think generally everyone around here is keenly focused on providing a strategy to confront what can only be described as an awesome challenge for the country.

5:29There are also some opportunities with AI, but the challenge of it is, I don't think I have to exaggerate that. Over in the world of cool tools, people are really excited about the latest from Runway. Runway is the generative video creator that has evolved from a video-to-video modifier to a text-to-video creator to now something that can generate video from a starting image. Basically, you can take an image that you've created in something like Mid Journey or Stable Diffusion, and then Runway Gen 2 will actually generate a video around it. The last thing I want to mention in today's brief is that ChatGPT has rolled out a new feature called Custom Instructions.

6:03I'm going to do an entire video for this this weekend, which will likely be out on Sunday, but Sam Altman described Custom Instructions as one small step towards more personalized AI, but a surprisingly fun and useful one. Custom Instructions effectively comes down to two questions. What would you like ChatGPT to know about you to provide better responses? And how would you like ChatGPT to respond? The important thing is that you will not have to rewrite this with every prompt. Instead, ChatGPT will remember these instructions and respond accordingly. There is a ton of excitement about this feature on Twitter, so if you want to hear more about how people are using that, make sure you subscribe to the channel and come back for more of that soon.

6:41Anyways guys, that's going to do it for today's AI Breakdown Brief. If you're enjoying this but would find it useful to get an email form, go subscribe to the AI Breakdown newsletter. There's a link to that on breakdown.network. Thanks for listening or watching as always, and I'll be back soon with the main AI breakdown. Before we get to the main episode, I want to tell you about today's sponsor, NetSuite. I know many of you guys are entrepreneurs, executives, managers, business leaders who are trying to figure out how technology is changing the world and how it can change your business. Given that, I am thrilled to have NetSuite as a sponsor of the AI Breakdown.

7:17NetSuite gives you the visibility and control you need to make better decisions faster. It is the software superpower behind so many of the world's most successful companies. And for the first time in NetSuite's 25 years as the number one cloud financial system, you can defer payments of a full NetSuite implementation for six months. That's no payment and no interest for six months, and you can take advantage of this special financing offer today. Listen, NetSuite is number one because they give your business everything you need in real time, all in one place to reduce manual processes, boost efficiency, build forecasts, and increase productivity across every department.

7:52I know that if you are listening to the AI breakdown, you understand intuitively and deeply just how much data matters to any modern business. Having all of your information in one place can be the difference between making the right decision and making the wrong one. I think it's awesome that NetSuite has this new offer designed to really make their suite of tools available for all the businesses that need it. So, if you have been sizing up NetSuite to make the switch, then you know that this deal is unprecedented. No interest, no payments. Take advantage of this special financing offer at netsuite.com slash breakdown.

8:25Go to netsuite.com slash breakdown to get visibility and control you need to weather any storm. That's netsuite.com slash breakdown. And with that, let's get to the show. Welcome back to the AI Breakdown. For the last several months, the White House has been increasingly interested in and engaged with artificial intelligence. At the beginning of May, Vice President Kamala Harris met with a number of CEOs to discuss how they could collaborate with regulators to move AI forward in a way that was taking advantage of its benefits, but also recognizing the risks and challenges it presented to society and the economy.

8:59And part of the initiatives that they announced around that was that they were going to be working with a number of leading AI companies to try to come together around a set of voluntary commitments. The goal of this wasn't to replace congressional legislation, but simply to try to keep things moving and come to some agreement with industry around some common sense measures that could be implemented more quickly. Today, Friday, July 21st, the White House released a fact sheet. The title reads, Biden-Harris administration secures voluntary commitments from leading artificial intelligence companies to manage the risks posed by AI.

9:29The goal with these voluntary commitments is to help the companies involved move toward, quote, safe, secure, and transparent development of AI technology. These commitments, they write, which the companies have chosen to undertake immediately, underscore three principles that must be fundamental to the future of AI, safety, security, and trust, and mark a critical step towards developing responsible AI. As the pace of innovation continues to accelerate, the Biden-Harris administration will continue to remind these companies of their responsibilities and take decisive action to keep Americans safe.

9:59So first of all, let's talk about the companies involved. They include Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI. Now, this is kind of the companies that you would expect. From a big tech perspective, Apple is absent, but then again, they're keeping their AI approach close to the vest, even if reporting is starting to leak out about it. One potentially interesting participant is Amazon, who so far as well have been a little bit circumspect about their long-term plans with AI, even as they've released things like a developer sandbox for people to use AI models integrated directly into Amazon Web Services.

10:30So what are the commitments that these companies are making? The first category relates to ensuring products are safe before introducing them to the public. First, the companies commit to internal and external security testing of their AI systems before their release. The testing, which will be carried out in part by independent experts, guards against some of the most significant sources of AI risks, such as biosecurity and cybersecurity, as well as its broader societal effects. Second, the companies commit to sharing information across the industry and with governments, civil society, and academia on managing AI risks.

10:59This includes best practices for safety, information on attempts to circumvent safeguards, and technical collaboration. So you can see here one of the key themes is to not let the development of these technologies be complete black boxes. The administration wants there to be more eyes on any given system before it gets to the public. And what's more, when a company learns about how to mitigate a risk, they want that knowledge to flow freely. The next category is building systems that put security first. The first commitment in that area is that the companies commit to investing in cybersecurity and insider threat safeguards to protect proprietary and unreleased model weights.

11:34These model weights are the most essential part of an AI system, and the companies agree that it is vital that the model weights be released only when intended and when security risks are considered. One of the big points of discussion earlier this week when Meta announced Lama 2 is that if they released the model weights, then some people felt that all their safety measures were for naught because people with malicious intent could simply fine-tune around them. The second security first commitment is that the companies commit to facilitating third-party discovery and reporting of vulnerabilities in their AI systems.

12:03And this basically says that that external testing happens even after a model is live in the wild. Quote, Some issues may persist even after an AI system is released, and a robust reporting mechanism enables them to be found and fixed quickly. The third category of commitments is around what the White House calls earning the public's trust. First, the companies commit to developing robust technical mechanisms to ensure that users know when content is AI generated, such as a watermarking system. This action enables creativity with AI to flourish, but reduces the dangers of fraud and deception. Broadly speaking, this is one of the things that I think there is the most agreement around, that we need ways to allow people to differentiate what is generated by AI versus real in the way that we classically understand that word today.

12:43Next, the companies commit to publicly reporting their AI systems' capabilities, limitations, and areas of appropriate and inappropriate use. This report will cover both security risks and societal risks, such as the effects on fairness and bias. Third, the companies commit to prioritizing research on the societal risks that AI systems can pose, including on avoiding harmful bias and discrimination and protecting privacy. The track record of AI shows the insidiousness and prevalence of these dangers, and the companies commit to rolling out AI that mitigates them. One of the things that has been really clear in the White House's engagement in particular with this is that while there may occasionally be some mention of those bigger, broader risks, there's definitely more of an emphasis on the stuff that is politically hot-button right now.

13:23There is a great concern that if the information that is training these systems has inherent biases, then those biases will be passed on and even amplified in AI systems. This is also part and parcel of why you're seeing legislation being introduced, for example, to prevent employers from using entirely AI systems for making hiring decisions. There's just too much of a concern that those systems reflect the worst, not the best parts of their training. Finally, to sort of wrap it up with a nice bow commitment, the companies commit to develop and deploy advanced AI systems to help address society's greatest challenges.

13:54From cancer prevention to mitigating climate change to so much in between, AI, if properly managed can contribute enormously to the prosperity, equality, and security of all. This is obviously what these companies were already doing, and I think the reason it's here is to reflect the fact that the White House is trying to send a signal that they don't want to prevent the development of this technology. They want to create guardrails that prevent the biggest part of the harms so that its potential benefits, enumerated here, can come to the fore. Now, as part of this announcement, the White House also said that it had been consulting with a huge array of countries on similar types of voluntary commitments.

14:27Those countries include Australia, Brazil, Canada, Chile, France, Germany, India, Israel, Italy, Japan, Kenya, Mexico, the Netherlands, New Zealand, Nigeria, the Philippines, Singapore, South Korea, UAE, and the UK. So I think the big question and the big skepticism is whether industry commitments really matter at all. The cynical take is that these are a way around other more strenuous regulation, and you see a lot of people on Twitter saying things like the EU is much stricter. The reality, of course, is that the EU AI Act and these voluntary commitments aren't really comparable things. One is legislation from an elected body.

14:59The other is a voluntary set of norms agreed upon by the leaders of an industry. A New York Times piece about the announcement recognizes that if nothing else, this type of commitment could prevent a countervailing force to the endless arms race that, from a financial and markets perspective, is gripping all these companies right now. Now, other things that people don't love.

15:32On the other side, some are finding that this sort of proactive and productive engagement with industry is a welcome switch from a White House which has seen in other areas a lot of contentiousness in that relationship. Wedbush analyst Dan Ives writes, and like the self-regulatory strategy for big tech in this AI arms race. As we have seen with black eye after black eye from Lena Kahn and the FTC against big tech, this is the better route with AI in the US in our opinion. The companies involved have also released a number of statements. Microsoft's Brad Smith writes, By moving quickly, the White House's commitments create a foundation to help ensure the promise of AI stays ahead of its risks.

16:08We welcome the president's leadership in bringing the tech industry together to hammer out concrete steps that will make AI safer, more secure, and more beneficial for the public. Mustafa Suleiman writes,

16:46today that will include the participation of all these companies, including Brad Smith from Microsoft, Nick Clegg, the head of global affairs for Meta, Ken Walker, the head of global affairs for Google, Greg Brockman, a co-founder at OpenAI, Adam Solipsky of Amazon Web Services, Dario Amodi of Anthropic, and Mustafa Suleiman of Inflection AI. The TLDR for me on things like this is that they are ultimately more signaling than anything else. This is not meant, I don't believe to take the place of real regulation. It's not meant even really to take the place of broader industry self-regulation. Instead, in some ways, this is closer to a more proactive version of signing a letter that says this type of approach to AI development is important, and we want to be a part of the conversation around how to do that well.

17:30I think when put in that context, it is a small but positive and meaningful step. Of course, it doesn't solve the big challenges of AI, but it wasn't meant to. And frankly, right now, anything that gets this set of actors cooperating and remembering that they have more incentives than just a commercial incentive to race past one another is, I think, a positive thing. If there's anything interesting that comes out of the White House meeting later today, you can be assured that I will tell you guys all about it next week. For now, I want to say thanks one more time for listening to the AI Breakdown, and until next time, peace.

18:07Thank you.