Google researchers introduced Auto-Diagnose, an LLM-powered tool that reads failure logs from broken integration tests, identifies the root cause, and posts a concise diagnosis directly into the code review where the failure appeared. In a manual evaluation of 71 real-world failures spanning 39 teams, the tool correctly identified the root cause 90.14% of the time. It has run on 52,635 distinct failing tests across 224,782 executions. The system aims to reduce time spent sifting logs and shorten debugging cycles. The tool integrates with existing code review workflows, surfacing diagnoses where failures were detected.

• Identify root causes from failure logs
• Post diagnosis directly into code reviews
• Achieve 90.14% accuracy on 71 real-world failures
• Scale across thousands of tests
• Integrate with CI/CD workflows

KV cache can increase VRAM usage during model inference as K and V matrices are recomputed for previous tokens. The article explains why this memory burden occurs in attention mechanisms and presents TurboQuant as a solution to address the issue. TurboQuant aims to optimize memory usage during autoregressive inference, potentially reducing GPU memory footprints for longer sequences. The piece discusses practical implications for Transformer-based pipelines and how this approach could be adopted across models and deployments.

• Explain KV cache memory cost during inference
• Highlight TurboQuant as a memory-optimization solution
• Illustrate potential memory reductions for long sequences
• Encourage integration into Transformer inference pipelines

Five artist-engineer teams presented prototypes at the AI & Africa Music (AIAM) Project showcase at the Chris Seabrooke Music Hall, Wits, capping six months of collaboration on how AI can preserve, reimagine, and responsibly co-create with African music practices. Led by Professor Christo Doherty from the Wits Innovation Centre (WIC) in partnership with the MIND Institute, the initiative united academics, students and representatives from the creative, music, and technology sectors. Supported by Wits alumnus and US-based music executive Charles Goldstuck, the five winning projects highlighted issues including ethics, consent, provenance and creative ownership. The event underscored Africa-led leadership in AI music and the importance of ethical and inclusive co-creation with local musical traditions.

• Showcased five winning prototypes
• Demonstrated cross-sector collaboration
• Addressed ethics, consent, provenance and ownership
• Highlighted Africa's leadership in AI music