Developer tooling vendor Replit, renowned for its cloud software development platform, has embarked on a mission to democratize artificial intelligence for developers worldwide. With a user base exceeding 20 million individuals, Replit's latest strategic moves aim to bring AI capabilities to the fingertips of every developer, marking a significant shift in their approach to AI integration.
Over the past year, Replit has been progressively enhancing its generative AI offerings, a journey that commenced with the introduction of the GhostWriter AI code completion tool and a strategic partnership with tech giant Google. However, access to GhostWriter had been limited to a subset of Replit users. This, Replit is about to change. Effective October 9, Replit GhostWriter makes the generative AI code completion tool available to all of its users, an initiative aptly named "AI for All."
In conjunction with the GhostWriter integration, Replit is also unveiling a new version of its open-source generative AI LLM for coding, named replit-code-v1.5-3b. This open-source LLM is positioned as a formidable alternative to competitors like StarCoder LLM, a collaborative effort between ServiceNow and Hugging Face, and Meta's Llama CodeLlama 7B.
Amjad Masad, the CEO of Replit, underlined the company's mission of providing equitable access during a live-streamed session at the AI Engineer Summit. "Our mission is to empower the next billion developers, and so we really didn't want to create this world where some people have access to AI GhostWriter and other people don't have access to it," Masad stated.
In recent years, the tech world has witnessed a proliferation of AI-powered coding tools designed to facilitate and enhance developers' coding experiences. Major players like Microsoft with GitHub Co-pilot, Amazon with Codewhisper, and even the Stack Overflow platform with its OverFlow AI project have entered this domain. However, Replit's vision differs from that of its competitors. Masad believes that AI should not be an add-on but an inherent part of the coding experience.
With this integration, Replit is not only doing away with the GhostWriter label but is also making AI an integral, default feature for all its users. Masad stressed that Replit's diverse user base includes people from around the world coding on various devices, including laptops and mobile phones. Now, all of these users can become AI-enhanced developers.
"We think this is going to be the biggest deployment of AI-enhanced coding in the world," Masad boldly claimed. "We're going to be burning as much GPU as we're burning CPU, so pray for us."
Crucially, Replit's generative AI capabilities are not merely wrappers on top of another vendor's LLM. They are constructed on proprietary open-source technology developed in-house. Michele Catasta, VP of AI at Replit, explained during the AI Engineer Summit that their code completion feature is powered by a bespoke large language model.
Replit's LLM underwent a significant upgrade, transitioning from replit-code-v1-3b to replit-code-v1.5-3b. This update significantly enhances the model's capabilities, and Catasta mentioned that it was trained on a staggering 1 trillion tokens of code, boasting support for 30 different programming languages.
Catasta emphasized the "secret ingredient" in the new LLM, which was the meticulous work dedicated to data quality. This has played a pivotal role in enhancing the model's performance. Furthermore, Replit's investment in powerful hardware is evident, as the LLM was trained on 128 Nvidia H100-80G GPUs, a resource as precious as gold. Catasta proudly declared that this is the first model officially announced to be trained on the H100 that has been released as open source, marking a significant milestone in the AI journey of Replit.