ManufacturingAutocoding platforms have emerged as powerful tools for developers, employing large language models to generate code from natural language prompts. While these platforms offer great promise, concerns regarding their use of data have also arisen. For example, Microsoft has faced lawsuits for using GitHub repositories to train its Copilot algorithm without crediting the original developers. Amid these challenges, some autocoding platforms, such as Tabnine, prioritize responsibly-sourced datasets and user privacy, setting them apart from competitors like GitHub Copilot.
In an interview with Analytics India Magazine, Brandon Jung, VP Ecosystem and Business Development at Tabnine, highlights the benefits and challenges associated with responsibly-sourced datasets. According to Jung, relying on fully permissive licensed code can limit the amount of data available for training. However, this approach prevents the inclusion of low-quality or non-permissive code, as well as personal information. By committing to using only fully permissive code, Tabnine avoids unintended consequences and potential legal issues.
Despite the reduced dataset size, Jung contends that Tabnine can compete with GitHub Copilot. He points out that training on all of GitHub's data can bias the model toward outdated code and practices. In contrast, Tabnine's approach, which focuses on more recent, high-quality code, is more forward-looking. By collaborating with industry partners like Google and Amazon, Tabnine's datasets are oriented towards current APIs and reflect where the industry is headed, rather than where it has been.
