AI assistants come to Alibaba and a16z-backed cider

The world of generative AI is evolving so rapidly that every few days we see startups deploying new applications powered by Large Language Models (LLMs). The latest attempt to monetize artificial intelligence comes from Mindverse AIa Singapore startup building an API interface, or what founder Fangbo Tao calls a “base layer” for enterprises, to create intelligent agents with their own vertical memory and different skill sets using OpenAI’s GPT-series LLMs .

AI agents similar to Mindverse’s ChatGPT have already secured early adopters, including an undisclosed platform within the Alibaba ecosystem; trendy a16z-backed startup Cider, which is testing the virtual assistant; and Hooked, a web3 educational platform that leverages the startup’s AI agent to guide users through its site.

Given its traction and investor enthusiasm around conversational AI, it’s no surprise that Mindverse is close to completing a $10 million Series A funding round. Investors are likely to be easy on Tao’s experience working on AI systems at tech giants in China and the US. After a stint at Facebook building its content understanding platform, Tao joined Alibaba in Hangzhou to help found an internal AI lab before starting your own company.

Mindverse’s latest round, which raised $7 million, valued it at $45 million and was led by Sequoia China with participation from Linear Capital, K2 Venture, Yinxinggu Capital, and Plug and Play.

Mindverse essentially provides a platform that allows customers to quickly create specialized intelligent agents for different domains. This is what happens when a user lands on a Mindverse-powered eCommerce site: they’ll be greeted by a chatbot that has sucked in all of the site’s inventory data. Suppose the shopper asks something like, “What do I wear to my beach vacation?” The bot will search for the products and display some options.

By conversing in a human way, the purchasing agent can also explain product differences and suggest more alternatives if the user is not satisfied with their first recommendations, which means the bot can learn from conversations in real time.

Similarly, a hotel booking site can use Mindverse to create a virtual guide that recommends places to stay based on a simple input like “I’m planning a trip to San Francisco with my wife.” The locations shown will take into account the interests of both husband and wife rather than universal tourist hotspots.

This way of interacting with web data, Tao said, is fundamentally different from the pregenerative AI era.

“In the past, users interacted with data sources through software and applications, or GUIs. [graphical user interface]. What we are doing now is adding an agent or co-pilot to help the GUI… by training the AI ​​to autonomously learn the API, documents, data sources, and instructions that we provide so that the agent can acquire skills specific to business scenarios and provide dynamic orchestration. from those based on complex user intent,” she explained.

“The biggest difference is that existing recommendation algorithms rely heavily on data from the past and you can’t specify your needs,” he continued. “What you click or buy determines what you see. Through [generative AI]on the other hand, you can have an active back-and-forth interaction with the AI ​​agent that can digest your intent.”

However, it does not mean that recommendation algorithms will become obsolete. In fact, Mindverse agents can compare your recommendations against those of algorithms that learn from past data. One way to integrate both solutions is to embed the old algorithms into the agent as an API, so that the application can learn from past user behavior. In fact, any conventional capabilities behind the software, beyond recommendation and search, can be integrated into AI agents as API skills, the founder noted.

“But the AI ​​agent acts on a higher level. By chatting with users, you can make better use of the recommendation and search capabilities to plan how best to use the back-end data,” Tao said.