ainewsblitz.com

Breaking

Google DeepMind VP Says Coding Models Can Train Themselves as Human GitHub Data Runs Out

Google DeepMind VP Says Coding Models Can Train Themselves as Human GitHub Data Runs Out
  • Software Dev & Coding
  • Foundation Models
  • AI Agents

Google DeepMind vice president Benoit Schillings told developers that coding models can now improve by inventing and grading their own tasks, because the supply of human-written training examples is drying up. He said about 80% of new code added to GitHub is already machine-generated, and that further gains will come from AlphaZero-style self-play rather than more scraped human repositories.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 27,697 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year