Proving Claude Knows You by Your Code Style: Author2Vec (author2vec.com)

A new model called Author2Vec has demonstrated the ability to identify programmers based on their coding style, even from unseen code snippets. When trained on samples from 13 developers, the model exhibits improved recognition accuracy with increased exposure to a coder's previous work. In tests, guess rates started at 7.7% for unfamiliar code but soared as the model consumed more examples of an individual's coding, illustrating a clear gradient of recognition. This capability is significant for the AI/ML community as it opens doors to advanced code analysis, which can enhance code review processes, personalized learning, and even detection of code plagiarism. The underlying technique relies on vector embeddings that capture distinct stylistic traits of coders. When faced with entirely new passages, the model resorts to identifying stylistic similarities with known coders, highlighting how individual coding styles can be distilled into unique vector representations. As AI continues to evolve, such nuanced capabilities could revolutionize how we approach software development and digital collaboration.