Generative AI startups are attracting billions of dollars, but they are already doomed to failure if they don’t get the right data, and that’s no easy feat.
Generative AI startups are attracting billions of dollars, but they are already doomed to failure if they don’t get the right data, and that’s no easy feat.
“We’ve seen a lot of pitches from companies that seem to be pursuing great applications of AI, but they don’t have access to the data to build powerful applications, let alone They also didn’t have access to the proprietary data that backs them.They have a competitive moat in their business,” said Brad Svrluga, co-founder and general partner of venture capital firm Primary Venture Partners. increase.
Hi!reading premium articles
“We’ve seen a lot of pitches from companies that seem to be pursuing great applications of AI, but they don’t have access to the data to build powerful applications, let alone They also didn’t have access to the proprietary data that backs them.They have a competitive moat in their business,” said Brad Svrluga, co-founder and general partner of venture capital firm Primary Venture Partners. increase.
Today, having the right data is more important than ever to your success. Building the actual model is now somewhat commoditized, but the real value is in the data, said Paul Taima, resident chief technology officer at Bullpen Capital.
Venture funding for generative AI startups increased from $4.8 billion in 2022 to $12.7 billion in the first five months of 2023, according to PitchBook. Many of these companies are now looking to build more niche AI models in fields such as finance and healthcare, but accessing training data sets there has been a challenge.
Some AI startups are looking to partner with data-rich big companies. For example, Myrna Licker, EY Global Vice Chairman of Tax, said the company is approached daily by generative AI start-ups thanks to its vast amount of transactional data. But Andy Baldwin, global managing partner of EY’s client services, said he was concerned about what would happen if EY’s data were used to train external models.
“Who owns that data? What happens to access to that model when we train it? And how else can others use it?” says Baldwin. “Data is part of the intellectual property we serve.”
Startups can get around the IP problem by training different models for each client only on that client’s data. This is the strategy his TermSheet at the startup uses to build generative AI models to answer industry questions for its Ethan products, real estate developers, brokers and investors. But getting clients to agree also requires some education and persuasion, said CEO and co-founder Roger Smith.
Andy Wilson, co-founder and CEO of legal tech firm Logikcull, said that with a strong cybersecurity posture, convincing companies that they can actually protect their data can also be difficult. increase.
Primary Venture Partners’ Svrluga says large tech incumbents have an edge over startups in generative AI applications, partly because they already have the trust of large enterprise customers who are used to working with data. said it was possible.
Tracy Daniels, chief data officer at financial services firm The Trust, said it is currently looking at generative AI use cases only with big tech vendors, not startups. She said she can trust big vendors to keep her data safe.
This means that even startups that can get a head start on public data face the challenge of fleshing out their models with enterprise data sets. His Veesual, an AI startup that can generate images that look like people trying on clothes, initially leveraged publicly available images from the internet for training, but handed over data to enhance its models. We struggled to get the big retailers to agree with us.
CEO and co-founder Maxim Patt said in some cases, major retailers demanded huge payments or even stake in the company in exchange for how Veesual would benefit from its data. The transaction did not materialize.
PatentPal, a generative AI startup that helps law firms draft patent applications, has received training on publicly available patent applications, according to CEO and founder Jack Xu. He said there is an opportunity to improve the accuracy of the tool by continuing to train it on real customer feedback, encrypted or anonymized. But the task is complicated by the need to separate that feedback from sensitive and confidential data, including trade secrets.
“For early-stage startups, there are brand awareness issues, social proof issues,” he says.
But there is also pressure. Adam Strzok, founder and managing partner of Strzok Capital, said some start-ups are competing with each other to secure more data in specific niches and do it faster. said that
“If you believe you have your own dataset, you’re going to get it before them and negotiate exclusivity.In that sense, it’s almost an arms race,” he said. rice field.
