AI Briefing: Enterprise AI has some growing up to do

Popular chatbots like ChatGPT and Gemini have captured headlines, but executives buying and selling enterprise AI hope to bring adoption past the nascent stage.

Last week, top AI providers debuted new business-focused updates for large language models. Enterprise-focused Cohere released a new LLM for Microsoft Azure and Oracle called R+ to help companies improve the accuracy and relevance of LLMs while also covering 10 languages and cutting costs.

Separately, OpenAI expanded capabilities for companies to fine-tune and customize models. And Amazon announced more enterprise tools through its addition of AI models from French firm Mistral.

Meanwhile, a new survey conducted by the enterprise-focused startup Writer AI — which in January released its own enterprise-focused LLM called Palmyra — indicates some lingering concerns over the development of enterprise solutions in AI. Accord to the survey, just 17% of the 500 executives polled said their internally built generative AI tools have excelled while another 61% reported accuracy issues.

Writer’s findings might sound convenient for a company offering its own private AI solutions to enterprise customers. However, it also illustrates the ongoing debate around whether companies should build their own AI tools or use tech from others.

While various forecasts predict big growth for various aspects of enterprise AI, the advertising industry also is expected to benefit. In a new report by Bloomberg Intelligence, ad spend driven by generative AI could grow from $4.6 billion in 2023 to $53.2 billion in 2027 and even further to $206 billion by 2032. Meanwhile, analysts also expect software revenue for consumers and e-commerce to grow from $995 million in 2023 to $45 billion in 2032 — all very bullish predictions.

At an AI event hosted by Bloomberg Intelligence on April 4, executives tasked with selling AI models and computing power said they’re just starting to see enterprise interest as AI models improve and computing power increases. But some believe that enterprise adoption of generative AI has barely scratched the surface.

“I don’t think that enterprise adoption has even begun,” said Brian Venturo, co-founder and chief strategy officer at CoreWeave, which provides cloud-based computing to companies. “When I see things like [people saying] enterprise apps are constrained right now, I have no idea what the use case even is. … Right now it’s kind of like, ‘I’m gonna go create a picture of a cat.’”

Last year was mostly focused on proof-of-concept trials without real deployment by enterprise customers, according to Neerav Kingsland, head of global accounts at Anthropic. However, he said that’s starting to change. LexisNexis is using Anthropic’s models for legal analysis, hedge funds like Bridge Water and Jane Street Capital are using it for financial analysis, and platforms like Slack are using LLMs to summarize chats.

Although some executives feel underwhelmed, others say they’re already seeing real reasons for building generative AI tools internally. Companies like Mastercard think the practice helps drive adoption across a global footprint with a range of customers. At the Bloomberg Intelligence event, Raj Seshadri, Mastercard’s president of data and services, said Mastercard can “develop at scale [and] deploy it in a more customized local way.”

“It’s a capability that we were uniquely able to build,” Seshadri said during a talk about generative AI and finance. “Many of [our] global peers work with us because they can use what we have, and do it at a lower cost with greater effectiveness.”

Balancing proper protections with enough experimenting has led companies to test within a guarded and sandboxed environment, according to Navrina Singh, founder and CEO of the AI governance platform Credo AI. From toxicity and hallucinations to IP leakage, she said enterprise risks within frontier AI models are far less understood, and that companies “don’t even know what we don’t know.”

“Not even because they’re closed but because these systems are performing in ways that we don’t understand,” said Singh. “Explainability and transparency of these systems becomes critical. And when you don’t know how they’re responding and the way they’re responding, we don’t know what can go wrong.”

Another thing that has held companies’ clients back from deploying enterprise-grade AI: worries about the accuracy and security of their data. Writer AI’s survey found that 95% of respondents thought more security measures are needed, while 94% saw data protection as a major concern.

“The public large language models in my view are kind of equivalent to search engines,” said Dell’s global CTO John Roese. “They’re extremely good at not much, but they do everything. And that makes them very interesting, much like search engines. Search engines are populated by information in the public domain. … If an enterprise was dumb enough to give up their data into one of those [public AI models], they’ve lost a gigantic competitive advantage.”

Prompts and Products: AI announcements and other news

  • Yahoo acquired Artifact, an AI news platform founded by Instagram’s co-founders that debuted last year.
  • Scott Donaton, who was previously a top marketer at Hulu, has joined the AI mobile content startup VersusGame as the company’s new CMO.
  • Amazon and the Hugging Face, an open-source AI platform, are going on a North American “roadshow” from April through early June to meet with developers.
  • Intel announced the fourth cohort for its startup accelerator program Intel Ignite.
  • In an interview with Bloomberg, YouTube’s CEO Neal Mohan said OpenAI might have violated the platform’s policies if YouTube content was used to train OpenAI’s text-to-video platform Sora.
  • A new report from eMarketer says marketers don’t think uses for generative AI are meeting their perceived potential.

Quotes from Humans: AppLovin’s Adam Foroughi on AI and social commerce

One of the biggest players in ad-tech is aiming to play a bigger role in the world of social commerce. Last week, AppLovin announced a $50 million strategic investment in Flip, a startup that lets users search and shop an online marketplace filled with influencer reviews and other user-generated content. (The funding is part of a $144 million Series C round raised by Flip.)

Through the deal, Flip will become the launch partner for AppLovin’s new ad marketplace for merchants. The partnership also gives AppLovin a way to play a bigger role in both social media and e-commerce. Powering the launch will be an updated version of AXON, AppLovin’s AI-powered adtech platform.

According to AppLovin co-founder and CEO Adam Foroughi, having AppLovin’s adtech power Flip will give AppLovin a way to expand its verticals, give merchants new ways to find shoppers and allow Flip to access AppLovin’s AI platform. He also sees it as a way for advertisers to reach users without third-party cookies. 

“It’s something that’s very, very appealing to the merchants we’ve talked to before,” he said. “And then to pair that with the Flip brand and the Flip ecosystem is also very appealing. You’ve got purchases happening directly in the app. It’s inside a mobile application outside of a browser, totally transactional, so it’s a closed loop. All the attribution is done within the Flip ecosystem.”

Other AI stories from across Digiday

https://digiday.com/?p=540157

More in Media

Publishers’ top tips for pitching advertisers at Cannes

The art of making sales at Cannes may not be the same strategy taught in business school.

AI Briefing: What would Steve Jobs say about Apple’s ambitions for AI?

Last week’s WWDC made plenty of headlines with various AI updates across Apple’s operating system and apps.

The Trade Desk and Yahoo are locked in a trading dispute with a looming deadline

If not resolved, it could result in the industry’s largest indie DSP cutting off access to Yahoo’s video inventory.