Insider Brief
- AI startup Symbolica raised $31 million in Series A funding led by Khosla Ventures, with participation from General Catalyst, Abstract Ventures, and Buckley Ventures.
- Symbolica aims to develop alternatives to the transformer deep learning architecture, which underpins current AI models like ChatGPT, and has co-written a paper with Google’s DeepMind.
- Symbolica’s first product will be a coding assistant, set to launch in early 2025, as the company focuses on hiring and training its model.
PRESS RELEASE — SAN FRANCISCO/April 9, 2024/(Reuters) — Artificial intelligence startup Symbolica, which develops foundation models to compete with ChatGPT creator OpenAI, announced a $31 million series A funding round led by Khosla Ventures on Tuesday.
General Catalyst, Abstract Ventures and Buckley Ventures also participated in the round, the company said.
Symbolica has created a framework that will enable it to develop alternatives to the “transformer” deep learning architecture, said Symbolica CEO George Morgan, who formerly worked at Tesla on its self-driving system, referencing a paper the company co-wrote with Google’s AI subsidiary DeepMind. Transformers are the basis for viral chatbot ChatGPT and the current generative AI race.
“The transformer is not the end-all-be-all of AI,” said Morgan. “What’s happening in the industry now is hacks on hacks on hacks.”
Whether or not AI can continue its rapid ascent in performance through pumping in more compute power and training data is hotly debated. Some companies, like Microsoft-backed OpenAI, are all-in on finding more compute. Others, like Symbolica, think different architectures of foundation models will be able to generate better results than scaling transformers.
“The industry broadly is contending with the limitations of transformers, whether it’s the costs required to scale them or their reliability,” General Catalyst partner Christopher Kauffman told Reuters.
Symbolica’s first product will be a coding assistant, although it will not launch until early 2025, Morgan said, as the company needs to hire and train its model, Morgan said.
(Reporting by Anna Tong in San Francisco; Editing by Stephen Coates)
