DeepMind’s AlphaCode research appears in Science on 8 December. The system was announced in February; today’s milestone is journal publication, not a product launch.
The researchers report an average top-54.3% ranking in simulated Codeforces contests. The ten contests postdate its training data. Their system generates millions of programs, then filters and clusters candidates into at most ten submissions per problem.
This measures contest problem-solving, not the full work of software engineering. The publication does not announce a public coding assistant.
DeepMind trademark; no endorsement.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.
