combat LLM spam by building a web of trust (blog.tangled.org)

Tangled has introduced a new "vouching" feature that allows users to endorse or denounce contributors, aiming to combat the influx of low-quality submissions generated by Large Language Model (LLM) tools. This feature assigns a green shield to vouched users and a red warning label to denounced users, creating a visual cue that helps users assess trustworthiness during interactions. The initiative is particularly significant for the AI/ML community, where the accessibility of LLMs has lowered the barrier for code contributions but also introduced the risk of subtle errors in submissions, increasing the burden on maintainers to review contributions carefully. The vouching process includes a text-based reason for the decision and operates within a network that only displays those decisions made by users and their immediate connections, promoting a focused web of trust. Currently, there are no penalties for denounced users beyond the visual indication of mistrust, which suggests a balanced approach that encourages engagement without harsh consequences. Future plans include features like vouch decay over time and evidence trails linked to contributions, enhancing transparency and accountability in the collaborative coding environment. This development emphasizes the importance of trust in online collaboration, particularly as AI tools become more prevalent in software development.