For all the pixels spilled about the promises of generative AI, it’s starting to feel like we’re telling the same story over and over again. AI is serviceable at document summarization and shows promise in customer service applications. But it generates fictions (the industry prefers the euphemistic and anthropomorphizing term “hallucinates”) and is limited by the data on which it’s trained.
Now that we’re at what seems like a plateau in the AI hype wave, where do startups and their investors see opportunities to use AI to help businesses operate in meaningful ways?
Venture capitalists we spoke with said efficiencies in internal workflows—whether coding, office-wide data search, or meeting and document summarization—were the biggest driving forces behind enterprise adoption right now.
“If you take a look at the macro, we’re kind of coming from a period of very rapid sort of Covid-driven growth for most businesses…that then turned into a bit of a headwind once the pandemic was over,” Ilya Fushman, a partner at Kleiner Perkins, said. “Those [headwinds] have driven a really strong need for efficiency.”
Low-hanging fruit
Fushman said AI tends to be more useful right now for applications that “don’t require 100% precision.” Things like text, video, and audio summarization, enterprise search, and entity extraction across data like customer conversations, documents, and messaging or video, he said.
“Those are kind of the obvious ones: Make people faster, especially in text-based settings,” Fushman said. “Most companies we talk to, the engineering teams are already using this, the design teams are using it, you know, customer-facing teams are starting to use it.”
Big companies like Goldman Sachs and AT&T, for instance, have already started to roll out generative AI to their developer teams.
Customer service is likely the next big area of experimentation as businesses use the tech to create chatbots to respond to consumer questions and problems, according to Sierra Ventures Managing Director Tim Guleri. But the use of AI in customer service remains fairly nascent; some companies may not yet feel comfortable with the tech powering full-fledged customer-facing chatbots.
“I think a thousand flowers can actually bloom just in the customer service realm, because it’s so inefficient, and over the decades, has been built up by effectively a lot of human capital applied to it,” Guleri said.
Sierra Ventures has raised $2.1 billion and made nearly 400 investments, per Crunchbase data.
More models needed
Anything beyond those relatively simple use cases, however, can require more complex types of models and structuring of systems, investors said. Guleri said improvements in AI’s ability to shuffle among text, voice, and imagery, and the chaining together of different large language models so that one’s input feeds into another, will ultimately expand the range of tasks that the technology can handle.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
“Some enterprises have very specific use cases, depending on the domain they’re in,” Guleri said. “And so a lot of the applications we have seen chain LLMs, and the chaining of the LLMs lets the applications use these LLMs in a very bespoke way for the business problem they’re trying to solve.”
Guleri said increased “verticalization of models” and smaller, specialized, and potentially open-source language models that can function without massive amounts of data to fuel them could open the door to a new evolution in generative AI.
“Once that starts happening, I think the floodgates are going to open on the value you can get,” Guleri said. “Because right now, we kind of have to sit around and twiddle our thumbs until OpenAI releases something.”
Nick Adams, co-founder and managing partner at Differential Ventures—which has raised $61.5 million in investment dollars and made 33 investments, according to Crunchbase—said the seed-stage enterprise data-focused fund is currently interested in smaller, specialized models that can operate with “a high level of accuracy and personalization.” Many of the ways that ChatGPT has been used most pervasively at the moment are “lower consequence” applications like drafting emails, creating marketing collateral, and conducting research, according to Adams.
“Then you have higher-consequence things that have real value to other parts of our society and US government interests, like healthcare and early detection of disease, drug discovery for medications,” Adams said. “Areas where we have dabbled quite a bit are AI for manufacturing to reduce costs and improve quality and safety in manufacturing, automating design of semiconductors and hardware components.”
A number of big companies and startups alike are exploring how AI might be used to automate the complex and arduous process of designing computer chips. Manufacturers are also experimenting with various ways the tech can be used in design and production processes, though some have noted that the inability to utilize intellectual property makes for datasets that aren't as expansive as they need to be to solve complex design challenges.
Overall, though, investors we spoke to remain bullish on the technology’s potential in the workplace despite its current limitations. Fushman said Kleiner Perkins—which has raised a total of $9.2 billion and made nearly 1,400 investments, according to Crunchbase—expects the technology to eventually pervade virtually every company.
“From our perspective, it’s very real. It’s a space that we keep a very close eye to,” Fushman said. “Essentially, every company is going to be an AI-powered company in some form or fashion, mostly because the infrastructure providers have done a really phenomenal job of making this technology very easily accessible.”