We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·26d ago·5 min read
GLM 5.2 is nearly as accurate as a human book keeper
TL;DR
GLM 5.2 was tested for preparing quarterly VAT returns for a UK SME, achieving nearly perfect accuracy. It processed 59 transactions in 68 minutes at a cost of just 2.73 USD.
✦ Why It Matters
Engineers can implement GLM 5.2 for automating VAT return preparation, reducing costs and improving accuracy for SMEs.
Key Takeaways
- Explore GLM 5.2 for financial automation tasks.
- Consider AI models for compliance-related processes.
- Evaluate cost savings in accounting with AI.
- Investigate the use of command-line tools for data input.
Full Summary
GLM 5.2, an open weights AI model, was evaluated for its ability to prepare quarterly value-added tax (VAT) returns for small and medium-sized enterprises (SMEs) in the UK. The benchmark involved processing transaction data and receipts from Vineyard Finance’s 2026 books, which were prepared by humans.
GLM 5.2 utilized a command-line interface (CLI) to input transactions into cloud-based accounting software, achieving a near-perfect VAT return with a discrepancy of only 7 pence (~10 US cents). The model completed the task in 68 minutes at a raw token cost of 2.73 USD.
This testing highlights the model's capability to handle compliance tasks traditionally performed by human accountants, suggesting significant cost savings for SMEs. The implications for engineers and researchers include exploring AI's potential in automating complex financial tasks.