AI for coding is still playing Go, not StarCraft
SMRTR summary
Artificial intelligence can now win gold medals at programming competitions, solving complex algorithmic puzzles with the precision of the world's smartest computer scientists. Yet these same systems stumble when faced with a messy, real-world codebase full of bugs and outdated documentation.
The disconnect mirrors what happened in gaming. AI conquered the ancient strategy game Go in 2016, but took three more years to master StarCraft 2, despite the video game being less intellectually prestigious. The difference? StarCraft's complexity and "fog of war" that hides information from players.
"There is so much to manage, there are so many different fights going on, there are so many little decisions you have to constantly make based on the information you have gathered," explains StarCraft commentator LowkoTV.
Software engineering resembles StarCraft more than Go. When production systems crash, engineers must navigate incomplete logs, conflicting library versions, and distributed failures across cloud infrastructure.
A recent study found developers using AI tools actually took longer to complete tasks in large projects, despite feeling more productive. We've reached the AlphaGo moment in coding, but we're nowhere near our AlphaStar breakthrough.
SMRTR provides this summary for quick context. The original article belongs to Daily.dev.
Read the original article