GLM 5.2 is nearly as accurate as a human book keeper (toot-books.com)

Article summary

GLM 5.2, an open weights AI model, was tested on preparing quarterly value-added tax (VAT) returns for a small UK business and achieved near-human accuracy at a significantly lower cost. The model processed 59 transactions in 68 minutes with a high degree of accuracy, although it made some mistakes, including a serious error in handling founding shares. The test results suggest that book-keeping is becoming a solved problem with the help of AI. The company behind the test is working on a solution to make this capability available to UK startups and SMEs.

Main themes

• AI in book-keeping
• VAT return preparation
• Accounting automation
• Cost savings
• Liability and risk
• Tax compliance

What commenters say

• The comparison to human book-keepers is relevant because it highlights the potential cost savings of using AI for book-keeping.
• The liability for errors remains with the user, even if an AI book-keeper is used, which could be a significant concern for businesses.
• Some commenters are skeptical about the reliability and trustworthiness of AI book-keepers, citing the potential for egregious errors and the need for human oversight.
• Others argue that AI book-keepers can be more accurate and efficient than human book-keepers, especially for routine tasks, and that the benefits outweigh the risks.
• The use of AI in book-keeping raises questions about the potential for tax authorities to use AI to audit tax returns, which could have significant implications for businesses and individuals.
• There are concerns about the security and privacy of using AI book-keepers, particularly if they require access to sensitive financial data.
• Some commenters suggest that using multiple AI models to cross-check results could help catch errors and improve accuracy.
• The distinction between book-keeping and tax preparation is important, and AI may be more suitable for certain tasks than others.