It’s the two year anniversary of ChatGPT and to ring it in, I found myself wandering through an enterprise software conference.
With a red AWS re:Invent lanyard around my neck last week, I moved past logos from companies like Deloitte, MongoDB and PwC, and wondered how a technology that once seemed poised to change the way we all interact with computers became the latest B2B software craze.
Outside of ChatGPT, consumer adoption of generative AI has been slow, but B2B firms have been pushing the tech to its limit. They are building agents, using AI to summarize massive amounts of documents, and fixing broken processes with models that make meaning from the madness.
Perhaps like the Blackberry, a business technology that demonstrated the power of the smartphone before the iPhone, enterprise AI will preview GenAI’s broader potential. So to learn where things are headed, I spent three packed days with the early adopters. Here’s what stood out:
Generative AI will be an enterprise thing for a while
Businesses have problems that people don’t. Most are trying to be hiveminds that centralize knowledge and act as a single organism. But so much business knowledge is contained within different systems, or kept by different people, or so voluminous that no human can efficiently consume it. So now, they’re embracing generative AI as a tool to ingest this knowledge, analyze it, and spit out the useful bits. The technology is well suited for their specific problem, and they’re investing heavily.
Matt Wood, an ex-Amazon executive who’s now the Commercial Technology and Innovation Officer at PwC, told me he believes that spending on generative AI today is split almost evenly between bots like ChatGPT and API access used to build enterprise tools. But “there’s much larger growth in the API usage” he said, speaking of the activity at PwC.
Enterprises’ embrace of generative AI should help fund the technology as consumer use cases get worked out. And because regular people don’t have a similar glaring problem that generative AI can solve, the “AI phone” and other consumer AI services aren’t likely to arrive imminently.
OpenAI and Amazon are on a collision course
If most of generative AI’s near-term growth is going to come via enterprise uses, then OpenAI needs to strike a deal with Amazon. After popularizing the technology, OpenAI is still behind in the enterprise game. And it still anticipates that most of its revenue will come from ChatGPT for the foreseeable future. Even if some of that growth is from ChatGPT Enterprise — which is used within companies — OpenAI needs more API use to grow and justify the $6.6 billion it just raised. And yes, it’s going to need to raise again relatively soon.
Enter Amazon. OpenAI’s foundational models are not accessible via the Amazon service that lets companies build with top AI models — called Bedrock — so it’s missing a major growth channel on the largest cloud service provider. If it’s going to take advantage of this moment of growth, it’s going to need to partner with Amazon in particular.
AWS CEO Matt Garman told me last week that he’s open to bringing OpenAI into the fold, so it’s now up to Sam Altman (and perhaps Satya Nadella) to make the deal. It seems inevitable, even if Amazon just released a new set of competitive AI models.
Agents are arriving (probably)
Moody’s, a rating agency that also offers analytics services, shared one of the most fascinating GenAI use cases I’ve heard so far. With a team of five people, the company built a swarm of agents that all do discreet tasks, but coordinate with each other to take on bigger projects. If someone is considering adding a company to their portfolio, for example, they could instruct a bunch of different AI agents to research one thing each. That could include the company’ financial status, where it’s located, macroeconomic conditions, industry trends, weather risks, and more.
When these agents complete their tasks, Moody’s then has them report back to other “collaborator” agents which collect their suggestions and run the results by “voting” agents that decide if the findings are reliable. Moody’s Chief Product Officer Nick Reed told me the company has 35 individual agents of this sort and gives them access to its proprietary data to help with their research. This idea of stacking agents (vs. relying on an individual bot) seems promising and could have uses beyond the enterprise (a system like this might be able to book travel?). We’ll see if they work as well as the marketing suggests, but the idea is fascinating.
Gen AI in healthcare is going to be big
Cancer patients often build up medical histories with thousands of pages of documents and physicians can struggle to read through all these documents and decide on the best course of action. Generative AI is already helping to solve this problem, giving doctors a summary of a patient’s medical history in a software program that GE Healthcare is currently running at Tampa General Hospital and UT Southwestern Medical Center.
When a doctor logs in, they get the patient’s summary view and see updates on their treatments, the cancer’s progress, whether the patient has missed a test, whether they qualify for a drug trial, and more. Parminder Bhatia, the Chief AI Officer of GE HealthCare, told me these models don’t hallucinate since they’re grounded in real patient data, and they also offer citations when doctors want to double check. You can imagine how useful this technology could be to the patients as well, giving them an opportunity to research and learn about their conditions, and steer their care.
This article is from Big Technology, a newsletter by Alex Kantrowitz.
The post Enterprise AI is Moving Fast, and Other Takeaways From a Week With Amazon appeared first on TheWrap.