Artificial Intelligence is highly popular today, especially text-generating AI or huge language models (imagine models similar to ChatGPT). Adopting big language models (LLMs) is viewed as a significant priority by 67.2% of enterprise firms surveyed recently (around 1,000). This poll was conducted in early 2024.
(Image Source: https://medium.com/)
However, obstacles are in the way. According to the same poll, many organizations could not deploy LLMs into production due to a lack of customization and flexibility and an inability to retain company expertise and intellectual property.
That made Varun Vummadi and Esha Manideep Dinne wonder what an enterprise LLM adoption solution may entail. Seeking one, they established Giga ML, a startup developing a platform enabling businesses to implement LLMs on-premise, purportedly saving costs and maintaining privacy.
Vummadi states, “Giga ML’s mission is to help enterprises deploy LLMs on their own on-premises infrastructure or virtual private cloud safely and efficiently.” “Giga ML takes care of the training, optimization, and operation of LLMs through an intuitive API, removing any associated hassle.”
For jobs like creating code and responding to frequently asked customer inquiries (such as “When can I expect my order to arrive?”), Giga ML offers its LLMs, the “X1 series.” The startup asserts that the models, based on Meta’s Llama 2, perform better than well-known LLMs on some benchmarks, especially the conversation MT-Bench test set. I couldn’t say how X1 compares in terms of quality; this reporter attempted to use Giga ML’s online demo but encountered technical difficulties. (No matter what I wrote in the question, the app timed out.)
After speaking with Vummadi, I thought Giga ML was more focused on developing tools enabling companies to optimize local LLMs rather than attempting to create the highest-performing LLMs available. This eliminates the need for reliance on external platforms and resources.
Vummadi highlighted the privacy benefits of offline models, which may persuade certain companies.
According to research by low-code AI development platform Predibase, less than 25% of businesses feel comfortable utilizing commercial LLMs due to worries about disclosing private or sensitive information to vendors. Because of concerns about privacy, expense, and lack of customization, about 77% of survey participants stated that they either don’t use or don’t plan to utilize commercial LLMs beyond prototypes in production.
Vummadi states, “IT managers at the C-suite level find Giga ML’s offerings valuable because of the fast inference, which ensures data compliance and maximum efficiency, and the customizable models tailored to their specific use case.”
With ~$3.74 million in venture capital funding raised thus far from Nexus Venture Partners, Y Combinator, Liquid 2 Ventures, 8vdx, and other investors, Giga ML intends to expand its two-person team and increase product research and development shortly. According to Vummadi, part of the funding also goes toward Giga ML’s clientele, which consists of unidentified “enterprise” firms in the financial and healthcare sectors.
Giga ML aims to enable companies to deploy Large Language Models (LLMs) offline, potentially revolutionizing privacy, security, and accessibility. This approach presents multiple benefits, including heightened data security, minimized latency, and functionality in internet-restricted environments.
Nevertheless, hurdles like managing updates, maintenance, and the setup process necessitate thoughtful solutions. Despite these challenges, Giga ML’s initiative could signify a transformative change in LLM deployment, paving the way for heightened privacy measures, enhanced efficiency, and broader applications of these advanced language models.
(Information Source: Techcrunch.com)