The AI coding assistant supplier Tabnine at this time introduced a brand new characteristic, Code Provenance and Attribution, to guard corporations from unintentionally including restrictively licensed code into their codebase when utilizing generative AI to jot down code.
The brand new characteristic checks AI-generated code towards public GitHub repositories to search out matches, and flags the license sort of the unique code from GitHub.
In keeping with Tabnine, this new characteristic will assist software program improvement groups perceive if the code generated by AI fashions meets their requirements and necessities.
“State-of-the-art LLMs like Claude 3.5 Sonnet and GPT-4o have tremendously improved the efficiency of generative AI purposes, together with AI code assistants. Nevertheless, these LLMs are skilled on huge quantities of knowledge collected from all corners of the web, together with code that will have restrictions on how it may be used, introducing the chance of IP infringement. For the reason that copyright legislation for the usage of AI-generated content material remains to be unsettled, engineering groups at enterprises need to strike a stability: leveraging the efficiency positive factors that come from these highly effective fashions whereas minimizing the chance of copyleft-licensed code getting of their codebase,” Tabnine wrote in a weblog publish.
Tabnine had already provided a license-compliant mannequin skilled solely on permissively licensed code, however this new characteristic will allow customers to leverage quite a lot of different fashions too, equivalent to Anthropic’s Claude, OpenAI’s GPT-4o, and Cohere’s Command R+.
The Code Provenance and Attribution functionality additionally helps different improvement actions on prime of code era inside Tabnine, together with fixing code, producing check instances, and implementing Jira points.
The corporate can also be engaged on increasing this functionality to allow customers to specify repos to verify towards, equivalent to a competitor’s code. It’ll additionally add a censorship functionality that removes matching code earlier than the developer sees it.
Code Provenance and Attribution is at the moment accessible as a non-public preview for all Tabnine Enterprise clients. Tabnine will even host a webinar on January 9 at 11 AM ET / 8 AM PT to dive into the potential additional.