AI 正着手打造更强的下一代 AI

TL;DR - AI systems are increasingly helping develop their successors through coding, architecture search, chip design, and automated research, but true recursive self-improvement remains unrealized. Human-defined goals, evaluation, oversight, and physical-world constraints still limit autonomous improvement loops.

• GPT-5.3-Codex, Claude Code, and AlphaEvolve illustrate AI’s growing role in software development, model optimization, and algorithm discovery.
• Darwin Gödel Machines can modify their agent code, while AI Scientist systems attempt to automate ideation, experimentation, writing, and review.
• Current systems still depend on humans to choose problems, define success criteria, verify outputs, and allocate resources.
• Researchers debate whether progress will produce rapid recursive improvement or slower “lossy” improvement constrained by system complexity, cost, and tacit human knowledge.