ChatGPT Enterprise Is Here - How It Shapes the AI Competitive Landscape

0:01Today on the AI Breakdown, we're looking at the announcement of ChatGPTBt Enterprise. Before that, on the brief, Elon and Zuckerberg go to Washington. 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 Discord, our newsletter, and our YouTube.

0:24Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. Now, today's big news is definitely, I think, ChatGPT launching their Enterprise Edition. This is a much-anticipated feature, and so the main episode will be all about that. But there were kind of a lot of other things that were announced recently as well that I think are also pretty meaningful. Let's kick off today with a policy discussion. Obviously, right now, all around the world, governments are trying to figure out how to make policy and how to regulate artificial intelligence. We did an episode last week talking about how governments are not only trying to figure out policy, but also trying to figure out how to leverage the fact that they're making policy as a competitive advantage.

1:04The United Kingdom is certainly positioning itself as a leader not only in policy, but also as being friendly to AI entrepreneurship. Spain just announced an AI agency last week that it's hoping to use in part, I think, to capture jobs. And then, of course, there is Washington, D.C. This year has seen meetings at the White House for AI leaders, and there's a growing conversation in Congress and the Senate about what it would mean to create guardrails around artificial intelligence, while also trying to understand that the U.S.'s leadership in the space is a competitive advantage not only in the economy, but also in geopolitics, especially vis-a-vis our relationship with China.

1:37Interestingly, relative to many new technologies, U.S. political leaders have been fairly open about the fact that they just don't know very much about this field and they need to catch up and get up to speed. I think that's a very productive disposition, and should it have been adopted in other areas, cough, cough, crypto, we might be in a much better place right now than we currently are. But in either case, one of the people who is pushing most is Senate Majority Leader Chuck Schumer. Earlier this summer, he promised to set up nine or more sessions for his colleagues in the Senate and in Washington more broadly to learn about and discuss key issues around artificial intelligence.

2:10On Monday, a spokesperson for Schumer said that the series of policy forums will kick off next month and that a group of very influential tech leaders will join as part of the first session. According to Schumer's office, attendees include Elon Musk, who, if you listened to yesterday's episode, has premiered new full self-driving for Tesla, which is powered by artificial intelligence. Mark Zuckerberg, of course, whose meta is one of the leaders particularly in the open-source-ish side of the AI space. And then in addition to those two, who of course still have not fought their cage fight yet, Microsoft's Satya Nadella, Alphabet's Sundar Pichai, OpenAI's Sam Altman, NVIDIA's Jensen Huang, and former Google CEO Eric Schmidt are all confirmed to be in attendance.

2:48The closed-door bipartisan meeting is to be held on September 13th. The spokesperson said that in addition to the tech leaders, it will also include representatives from advocacy, civil rights, workers, and creative groups. Obviously, the closer to that date we get, I think the more information we'll have, but that is a heck of a lot of firepower in one room at one time. And so you have to think that it might be more significant ultimately than just another photo op. Now, one of the reasons that politicians are feeling pressure to figure out some policy around artificial intelligence is that it appears to be growing as an issue that the electorate cares about.

3:19Pew Research just released a survey, and the announcement blog post they titled Growing Public Concern About the Role of Artificial Intelligence in Daily Life. The banner headline that has been splashed all around the internet is that 52 % of Americans say they feel more concerned than excited about the increased use of AI. Only 10 % said they are more excited than concerned, while 36 % said they feel an equal mix. A couple things about this that I think are interesting. One is that they've asked this question in each of the last two years as well. The numbers in 2021 and 22 were pretty consistent.

3:49In 2021, 37 % were more concerned than excited, whereas 38 % were more concerned than excited in 2022. It was 45 % and 46 % respectively who were equally excited and concerned, and 18 % and 15 % were more excited than concerned in those two years before this year. The rise of ChatGPT and all the other tools around it, the mid-journeys, etc. of the world, seem to have increase that anxiety, perhaps just with their display of power. Now, of course, the other possibility is that these results reflect the fact that media has been giving a ton of attention to the concern side of artificial intelligence, be it extinction risk concern or simply concern for jobs.

4:26We've also, of course, got the Hollywood strikes, which have AI as a central issue. So all in all, this kind of makes sense. Now, as you also might expect, the younger people are, the more excited they are, and the older they are, the more concerned they are, but it's still not totally dramatic. For example, of the 18 to 29 set, 42 % are more concerned versus excited, and 17 % are more excited versus concerned. That's certainly different than the 52 to 10 number overall, but not like an opposite or a huge shift. Now, interestingly, another really powerful part of this survey is when Americans were asked of their view of AI's impact on specific areas.

5:01When it came to keeping personal information private, Americans overwhelmingly said that AI hurts more than it helps. 53 % said it hurts more than it helps, versus only 10 % said that it helps more than it hurts. But in nearly every other area that they were asked, people said that AI helps more than it hurts. In terms of finding products and services online, 49 % said it helps more than it hurts, versus 15 % said that it hurts more than it helps. Companies making safe cars and trucks, 37 % said it helps more than it hurts, versus just 19 % said that it hurts more than it helps. Doctors providing quality care to patients, people taking care of their health, people finding accurate information online.

5:37All of these had more people thinking that it helps more than it hurts than the other way around, which I think adds credence to the idea that part of the concern may be on a very large general level. It may be big generalist concerns about jobs, about safety, or it may be that our concerns about privacy outweigh what we think are the benefits in these other specific areas. In either case, I think these are really interesting statistics. And since the survey was conducted between July 31st and August 6th, they're actually pretty up to date. Now, moving on to our next topic. As I mentioned, the big story today is ChatGPT Enterprise.

6:11But you'll remember that last week, the big update around OpenAI was that it was talking about its scraper that was going across the web and getting information for training future models, and that they told people how to block it. According to CNN Business, the number of major media and news organizations that have taken that proactive step to block chat GBT has gone up significantly. We heard about the New York Times and Reuters last week, but now others that have blocked GBTBot include Disney, Bloomberg, The Washington Post, The Atlantic, Axios, Insider, ABC News, ESPN, Kanye Nass, Hearst, and Vox.

6:40As CNN writes though, what exactly these media giants do next, however, remains to be seen. I think in the same way that the Hollywood strikes feel representative potentially of concerns more broadly about AI replacement for human labor, some of the legal fights around publishing and training data when it comes to authors and publications versus the big tech companies will likely have impacts on the shape of how AI is trained in the future as well. Overall, it continues to be super interesting times in AI land. If you are a person who likes consuming your news via the written word, let me end by asking you to check out the AI Breakdown newsletter.

7:15You can find a link on breakdown.network, which is obviously the network's website, or you can go to the AIbreakdown.beehive.com. That's B-E-E-H-I-I-V.com. I send out a newsletter with the five most important stories every morning, and I'd love to see you there. Thanks as always for listening or watching, and I'll be back soon with the main AI breakdown. Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage. If you work in a professional setting, you probably have some version of a one-on-one meeting, either with the people that work for you or the people that you work with.

7:48Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? That's where Supermanage comes in. Supermanage AI magically distills your team's public Slack channels into a real-time brief on any employee, any time. Catch up on contributions, work in progress, challenges they're facing, sentiment, everything you need to show up ready for a truly meaningful conversation, and it's completely free. Visit supermanage.ai forward slash breakdown today to start making the most of your one-on-ones. And thanks again to Supermanage for sponsoring the AI Breakdown.

8:24Welcome back to the AI Breakdown. Today, we are talking about the much-awaited announcement of ChatGPT Enterprise. We're going to talk about what the announcement actually has within it, some of the initial reactions, and what it means for the broader competitive landscape. So first of all, let's talk about the announcement itself. Sam Altman tweets, we launched chat GPT enterprise, enterprise-grade security and privacy, large-scale deployment support, unlimited and fast GPT-4, 32K context, and more. Customization on your company's data coming soon. So right out of the gate, we have some information about interesting things, some of which were expected, some of which were not, and some features which are not there yet, which could be consequential.

9:07OpenAI's announcement blog post reads,

9:22Now, of course, on top of those statements, they give a few more details. When it comes to what they call enterprise-grade security and privacy, OpenAI assures, quote, customer prompts and company data are not used for training OpenAI models. This is something that they announced when they first suggested that a ChatGPT business edition was coming, that it would be private by default. In fact, they announced this business model first in concert with an announcement that there was now a private no data training option even for the general ChatGPT. In addition, under enterprise grade security and privacy, they say that it's certified SOC 2 compliant.

9:55Now, this is an independent rating that's a voluntary certification from the American Institute of Certified Public Accountants that provides guidance on how organizations should manage customer data. There are five trust services criteria or elements within that, including security, availability, processing integrity, confidentiality, and privacy. But I think for the purposes of our conversation, what's relevant is that they are trying to say this is an independent type of certification that really does back up this claim that this is enterprise grade. Next up, they list their features for large-scale deployments.

10:25There is now an admin console with bulk member management. Obviously, if teams are deploying this across the enterprise or even across entire divisions, giving administrators the ability to change permissions, kick people out, all of that sort of stuff that you would expect from enterprise software, it sounds like that is now a part of this offering. There's single sign-on, which plugs into other systems, domain verification, and an analytics dashboard that'll provide admins the ability to understand usage patterns. Now, on top of that, they are also claiming that the enterprise edition provides access to the, most powerful version of chat GPT yet.

10:57Unlimited access to GPT-4, no usage caps. Of course, even if you are a chat GPT Plus user right now, there are caps on how much you can use GPT-4 versus GPT-3.5. There is faster inference or what they call higher speed performance of GPT-4, which they say is up to 2x faster. Unlimited access to advanced data analysis. This is interesting in both substance and marketing. They write this was formerly known as code interpreter. Now, obviously, code interpreter is one of the most significant features that OpenAI has ever released. About a month ago, I had a conversation with the guys from the Latent Space podcast about why code interpreter is such a significant upgrade that it effectively turns GPT-4 into GPT-4.5.

11:37The TLDR on that is that by giving ChatGPT the ability to write scripts to solve problems, it opens up a whole range of use cases that without code interpreter, it doesn't perform very well on, but with Code Interpreter, it can do really well. It makes total sense then that they are including this feature as a core part of the enterprise offering. And given how badly named Code Interpreter was, it also makes sense that they're moving towards the more descriptive explainer of advanced data analysis. The jump up to a 32k token context window is obviously a significant move. This allows for much larger enterprise documents, PDFs, etc.

12:13to be input without having to be broken up or without having to use things like embeddings. The Enterprise Edition also comes with credits to use their APIs for companies that find that they need a fully customized solution. And finally, they also have shareable chat templates for the company. So to the extent that there are common workflows, people don't have to do the same work over and over. I think in a lot of ways, this nails what you would expect from an enterprise offering from ChatGPT. Faster performance, code interpreter embedded, a bigger context window, GPT-4 all the time. All of this totally makes sense and makes it a good offering.

12:45And when it comes to demand, OpenAI says that it has been intense. They write, We've seen unprecedented demand for ChatGPT inside organizations. Since ChatGPT's launch just nine months ago, we've seen teams adopt it in over 80 % of Fortune 500 companies. Now that 80 % statistic refers to the percentage of Fortune 500 companies where someone using an email address associated with that corporate domain has signed up for ChatGPT. Obviously, that could be someone signing up to use it personally just with their corporate email address. But the inference that they're making is that if people are using their corporate email addresses to sign up, they're likely using it for work.

13:21OpenAI continues. We've heard from business leaders that they'd like a simple and safe way of deploying it in their organization. Early users of ChatGPT Enterprise, industry leaders like Block, Canva, Carlyle, the ST Lauder companies, PwC, and Zapier are redefining how they operate and are using ChatGPT to craft clear communications, accelerate coding tasks, rapidly explore answers to complex business questions, assist with creative work, and much more. So a couple things that are interesting from this. One, it's clear that there has been some beta period in which a number of different companies have been using this ChatGPT enterprise model or something close to it.

13:54And interestingly, those companies include not only tech companies that you might expect like Block and Canva, but also CPG companies like Estee Lauder. Now, what this brings up, and I think the most interesting question to explore around this is what the long-term pattern for enterprise adoption of generative AI, and specifically LLMs, is likely to be. One of my most referenced tweets from this year came from Sam Hogan, an AI entrepreneur, who in July wrote what was effectively a blog post but posted it to Twitter. He begins, six months ago, it looked like AI and LLMs were going to bring a much-needed revival to the venture startup ecosystem after a tough few years.

14:30With companies like Jasper starting to slow down, it's looking like this may not be the case. Now, this was nominally about how startups were not being as successful in the AI space as some might have expected. One of the reasons that Sam argued that was, is that the availability of open source tools, combined with the excitement of managers at enterprise companies, were leading many companies that might be natural customers of startups to spin up their own solutions using those open source tools, instead of working with a third-party startup that was untested, unproven, and might not be up to snuff when it comes to security, compliance, etc.

15:04Sam wrote, Executives at enterprise companies are excited about AI and have been vocal about this from the beginning. This led a lot of founders and VCs to believe these companies would make good first customers. What the startups building for these companies failed to realize is just how aligned and savvy executives and the engineers they manage would be at quickly getting AI into production using open-source tools. An engineering leader would rather spin up their own Langchain and Chroma infrastructure for free and build tech themselves than buy something from a new unproven startup. Now, when it comes to who was losing in this environment, Sam writes, Companies like Jasper and the VCs that back them are the biggest losers right now.

15:38Jasper raised over$100 million at a 10-figure valuation for what is essentially a generic thin wrapper around OpenAI. Their UX and brand are good but not great, and competition from companies building differentiated products specifically for high-value niches are making it very hard to grow with such a generic product. Now, obviously, this question of to what extent a company that is, as Sam put it, just a generic thin wrapper around ChatGPT or OpenAI can actually survive is a really important one. In June, when he was on his global tour, one of the things that it appeared that Sam Altman was doing in private meetings was reassuring some of the leading developers of the company that OpenAI's goal was not to compete with them across the full spectrum of use cases.

16:18Basically, if you're a startup or any company that's relying on someone else's API, you're subject to the whims of that company. And if the company that owns the API decides that they want to compete in your niche, there's not a lot you can do about it. A blog post which was later pulled that detailed Sam Altman's meetings with developers in London indicated that Altman said that in general, OpenAI was not interested in competing with its developers. Where they were likely to put some effort was in and around this enterprise or business use case, basically taking ChatGPT and customizing it for a business context.

16:50Given that, then, it probably shouldn't be a surprise that what we got was exactly that, and that companies who were effectively just offering an enterprise version of ChatGPT, taking advantage of the OpenAI APIs, are potentially in a bit of trouble. Jim Phan from NVIDIA writes, ChatGPT Enterprise, the beginning of the end of many B2B thin wrapper startups. Now, another interesting dimension of the competition question is OpenAI's increasingly complex relationship with Microsoft, its biggest investor. There has been a growing conversation about the extent to which these companies find themselves as collaborators and friends versus competitors and frenemies.

17:28Now, there's no denying that Microsoft has upside in OpenAI's success, but at the same time, that's certainly not stopping them or certainly seems not to be stopping them from diversifying their bets in the space. For example, Microsoft recently announced that it was planning to integrate a version of Databricks into its Azure suite, and Databricks is effectively a platform that helps enterprises create AI models from scratch or customize existing open source models and training them on their proprietary data. In that way, they represent an alternative to licensing OpenAI's models. Now, much hay has been made of this because it makes for a good story of Microsoft competing with OpenAI or there being some big shift in the relationship, but it might be as simple as a company the size of Microsoft not being willing to bet on only one horse and understanding or perhaps hearing from enterprise clients that part of what they want, given how important the proprietary data of a company is when it comes to some of the enterprise use cases of LLMs, that they might want to spin up their own solutions.

18:25Going back to Jim Phan's tweet again, he writes,

18:42Now Sam Altman seems to understand that this is a key priority, given that in his announcement tweet he wrote, Given that, it seems likely that OpenAI understands that while there are many use cases for which simply a more powerful version of ChatGPT that's customized for the enterprise makes sense. Indeed, they list some of them, right? Crafting clearer communications, accelerating coding tasks, etc. It may be that for some, what they really want out of an LLM is something that is trained from the ground up on their data. The big question is where on balance this ultimately lands. Is this actually a competition between this sort of third party model offered by OpenAI and ChatGPT Enterprise versus the customized model offered by something like Databricks or just building something from scratch using open source models?

19:29Or are they simply different tools based on the same technology but for different use cases? It wouldn't surprise me if the answer is the latter. For example, when it comes to crafting clearer communications, how much does having a model that's trained on or fine-tuned on one's corporate data really matter? One could argue that it does, that it gives the ability for the LLM to speak in the brand tone of the company in question, but it also may be that it's perfectly sufficient for the day-in, day-out use cases to just use the generic model. Ultimately, of course, the market will decide all these questions.

20:02And given how much the narrative has been on the idea that the model is going to be these enterprises who are customizing their own solutions versus just plugging into something like ChatGPT, I find myself somewhat skeptical. I do think in the long run that most companies will likely be fine-tuning big models on their data. However, I think in general, the pattern suggests that enterprise-customized solutions tend to fail in the face of more widely available third parties, and of course, more widely connected ecosystems. In 2012, David Strom wrote, whatever happened to intranets? He writes, back in the mid-1990s when the web was young, we had corporate intranets popping up all over the place.

20:42These were typically internal projects that were used to disseminate information to employees about projects, products, and customers. They were quick and dirty efforts that involved off-the-shelf parts and little, if any, programming. The idea was to produce a corporate web portal that was just for internal use, to enable staff to share documents, best practices, customer information, and the like. But they are mostly historical artifacts now. What happened? Well, for one thing, TCPIP happened. Back in the mid-90s, corporate networks were hodgepodge of protocols, including SNA and Netware. No one talks about these anymore.

21:09Having an all-IP network made it easier to adopt more internet-native technologies. Remember when sending emails from one company to another was a chore and not always successful? Now we take it for granted that we can communicate with anyone. Secondly, the tool sets got better. Many companies migrated their intranets to wikis or WordPress when it became clear that these products were easier to maintain and use. And then a whole class of products now called Enterprise Social Networks arrived, which have ready-made discussion groups, microblogs, news streams, and social media. The point being, ultimately, that the pattern of entrepreneurship and creativity ultimately favored open solutions and startups that were building custom solutions versus what engineering departments could spin up on their own.

21:45I do think that when it comes to AI, the value and importance of data and customizing data for a company is higher than in some of these just general communication use cases of the early internet. But I would be very surprised if at the end of the day, the fear of data leakage, the desire to train on one's own data from the ground up wins out against the convenience and the speed of iteration offered by external companies. That said, one thing that I think is very likely is that by and large, if the choice is between an unproven startup or an enterprise partner that already exists, a Microsoft, an Amazon Web Services, there are reasons to think that enterprises might favor those trusted partners already.

22:22But whatever happens, it's going to be interesting to see. And I think there are going to be a lot of companies out there who are very excited that ChatGPT Enterprise is now open for everyone. That's going to do it for today's AI Breakdown. If you enjoyed this, do me a favor and send it to someone who needs to go by ChatGPT Enterprise. Maybe this will help them figure out what the right solution for them is. Until next time, peace.