These APIs have a couple of problems: they are overly complex and costly. Probst explained, “OpenAI and similar closed-door language models are designed for general use, not specific use cases. Right now, they’re excessively trained and costly for specific needs.” He also added, “OpenAI will certainly thrive, but we believe most of the market will need to find their own solutions. This is why open source options are very attractive to them.”
How Open AI Will Not Be Universally Applicable
OpenAI’s CEO Sam Altman also thinks that AI models won’t be universally applicable. “I believe both play significant roles. We’re keen on both, and the future will combine aspects of both,” Altman commented when responding to a query about compact, specialized models versus comprehensive models during a Q&A session at Station F earlier this year.
AI in Transition: Evolving Regulations, Shifting Enterprises, and the Significance of MLOps
Ethical and legal concerns are also associated with AI utilization. Regulations are still evolving, but European legislation, in particular, may push companies to employ AI models trained with highly specific data sets and methods.
According to Gartner, 75% of enterprises will transition from proofs of concept to production in 2024. So, the upcoming year or two represent some of the most crucial moments in the history of AI, as we’re now moving into production, likely using a combination of open-source foundational models customized with proprietary data,” Tahir explained.
He further noted, “The significance of MLOps lies in our belief that 99% of AI applications will rely on more specialized, cost-effective, and smaller models trained in-house.”
