Eyebrow-raising seed rounds based on nothing but a pitch deck are becoming de rigueur in AI startup land this year — particularly when there are some big name companies on the founding team’s CVs.
Adaptive, a new Gen AI startup looking to raise $20m, has them in spades, according to a pitch deck seen by Sifted.
Its team includes several engineers and researchers hailing from French-founded open source AI platform Hugging Face — CEO Julien Launay, who’s based in the Netherlands, as well as Daniel Hesslow, Alessandro Cappelli and Guilherme Penedo — who are working on software and large language model (LLM) research. This team was also responsible for much of the work on Falcon, a popular open source LLM released by the United Arab Emirates’ Technology Innovation Institute.
Rounding out the group is also Olivier Cruchant, who previously worked in the machine learning team at Amazon Web Services. Sifted contacted the Adaptive team members for comment.
Index Ventures is in talks to lead the new round at a $100m valuation, The Information reported on Monday. The firm did not return Sifted’s request for comment.
While a $100m valuation for a startup that has only just got out of the gate might beggar belief, it’s further proof that being a member of the small crop of engineers and researchers that have built large language models continues to serve as catnip for VCs.
Adaptive’s pitch deck says the company is building solutions to help clients build more accurate and efficient specialised AI models based on company data, as well as user interactions and feedback. It says it will also enable teams to improve their AI models without needing any knowledge of techniques like reinforcement learning.
The Adaptive team says it’ll be able to charge customers $120k-360k in ARR for ongoing use of its platform, and up to roughly $1m for custom models in so-called “one-off revenue,” per the document.
Adaptive plans to use its $20m to hit a series of milestones in the next two years, including adding support for diffusion models (models like Stable Diffusion that are used for image generation) by the end of 2024.