Code up, large language models! Salesforce exec’s key advice. Level up.

Summary of Article


Former Salesforce Exec Says AI Must to Learn How to Code to Level Up

  • A former Salesforce executive, Richard Socher, believes that one way to improve AI is to make it program responses, rather than just predict them.
  • Large language models (LLMs) currently only predict the next token, but forcing them to code could give them more capabilities.

Key Points

Generative AI technology has advanced rapidly, but Richard Socher believes it still has further to go.

Socher suggests that forcing large language models to respond to certain prompts in code could level up their capabilities.

By translating questions into computer code, AI models are more likely to produce accurate answers, particularly for complex mathematical questions.

Scaling AI models by increasing data and computing power may reach its limits, according to Socher.

Socher’s views come as large language models continue to compete with OpenAI’s GPT-4 model.