GPT 4 - AshokBhat/ml GitHub Wiki

About

Highlights

  • Extension of text input to multimodal signals
  • Stronger capacity to solve complex tasks
  • Large performance improvement on evaluation tasks
  • Superior performance compared to prior GPT models
  • Safer response to malicious or provocative queries
  • Six-month iterative alignment process in RLHF training
  • Intervention strategies to mitigate LLM issues
  • Introduction of red teaming for reducing harmful content
  • Developed on a well-established deep-learning infrastructure
  • Improved optimization methods
  • Mechanism of predictable scaling for accurate performance prediction

Use-cases

  • Predicting text
  • Produce human-like text

See also