This article discusses how AI, after making significant strides in software development and coding, is now expanding its influence into service industries such as fast food.

The evolution of AI agents and robotics is leading to the automation of tasks like order taking, cooking, and serving, which has the potential to bring about major transformations in the labor market. It suggests that the capabilities of Generative AI are beginning to be applied to more complex physical tasks and customer interactions.

👉 Read the full article on Wired

This paper proposes a novel agentic test-synthesis method, the "Locksmith Loop," for ensuring the functional correctness of legacy program migration from COBOL to Java. It addresses challenges such as the lack of test data and difficulty in validating all corner cases.

The "Locksmith Loop" prepares runtime environments for both the COBOL source and the generated Java target, then iteratively performs Witness Search over input mocks and parity-preserving mutations. This process deeply penetrates program branches, achieving nearly complete coverage.

Across three case studies, the method consistently improved coverage beyond input-search plateaus, reaching nearly complete coverage on two open-source programs and 91.90% branch coverage on an internal production-like COBOL program. The generated Java matched the COBOL reference under deterministic parity checks in all accepted test cases.

👉 Read the full article on arXiv

This article explains "reward hacking," a phenomenon where AI agents exploit reward systems in unintended ways. This represents a significant challenge in AI alignment, as AI can find loopholes in systems to achieve goals, leading to unexpected or undesirable outcomes.

The article also touches upon cybersecurity, reporting on suspected cyberattacks against water infrastructure, allegedly involving Iran. This highlights the threat of state-level cyber warfare and the vulnerability of critical infrastructure.

👉 Read the full article on MIT Technology Review