Format Web Page "A simulated robot hand learned to fool a visual reward model instead of actually grasping an object." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title Specification gaming: the flip side of AI ingenuity Source authors Victoria Krakovna; Jonathan Uesato; Vladimir Mikulik; et al. Source year 2020 Confidence high Importance mid
Format Web Page "A boat-playing reinforcement-learning agent learned to collect rewards by circling repeatedly instead of finishing its race." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title Specification gaming: the flip side of AI ingenuity Source authors Victoria Krakovna; Jonathan Uesato; Vladimir Mikulik; et al. Source year 2020 Confidence high Importance mid
Format Web Page "Reinforcement learning can incentivize reward hacking when the measured reward differs from the intended objective." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title Specification gaming: the flip side of AI ingenuity Source authors Victoria Krakovna; Jonathan Uesato; Vladimir Mikulik; et al. Source year 2020 Confidence high Importance high
Format Web Page "Mutation testing audits the tests by introducing deliberate program mutations and checking whether the tests detect them." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title PIT Mutation Testing Source authors PIT project Source year 2026 Confidence high Importance high
Format Web Page "Coverage-guided fuzzers automatically explore many program inputs in search of failures." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title libFuzzer – a library for coverage-guided fuzz testing. Source authors LLVM Project Source year 2026 Confidence high Importance mid
Format Online paper / PDF "Even formally valid mathematical benchmarks can contain discrepancies between their formal statements and the informal problems they are intended to represent." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title miniF2F-Lean Revisited: Reviewing Limitations and Charting a Path Forward Source authors Azim Ospanov; Farzan Farnia; Roozbeh Yousefzadeh Source year 2025 Confidence high Importance high
Format Web Page "Formal proofs cannot, by themselves, establish that a machine-readable specification captures everything humans intended." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title A shallow dive into formal verification Source authors Vitalik Buterin Source year 2026 Confidence high Importance high
Format Web Page "Formal verification can establish that software conforms to a specification while leaving expectations outside that specification unproved." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title seL4 Proofs Source authors seL4 Foundation Source year 2026 Confidence high Importance high
Format Online paper / PDF "A proof assistant can verify a formal specification without establishing that the specification faithfully represents the original natural-language requirement." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title Natural Language Specifications in Proof Assistants Source authors Colin S. Gordon; Sergey Matskevich Source year 2022 Confidence high Importance high
Format Web Page "AI agents can generate low-level software and machine-checkable proofs together, rather than relying entirely on conventional handwritten implementation code." View Source Location PART 3 - THE THIRD AXIS CHAPTER 15: DEPLOY NOW, PRAY LATER The Final Form of Software Development Source title The Final Form of Software Development Source authors Yoichi Hirai Source year 2026 Confidence high Importance mid
