Sam Altman: OpenAI CEO on GPT-4, ChatGPT, and the Future of AI | Lex Fridman Podcast #367

Alignment

The alignment of AI models with human values is a complex process that Sam Altman and his team at OpenAI are deeply invested in. They have developed a combination of internal and external efforts to align AI models like GPT-4, aiming to increase alignment faster than capability progress 1. Altman emphasizes that alignment and capability are closely linked, with better alignment techniques leading to more capable models 1.

Better alignment techniques lead to better capabilities and vice versa.

--- Sam Altman

Transparency in AI models is another challenge, as Altman notes the pressure from public scrutiny and the need to balance transparency with the risk of misinterpretation 2.

   

Risks

AI deployment carries significant risks, including misinformation and misuse, which Lex Fridman and Sam discuss in detail. The System Card document released by OpenAI highlights efforts to mitigate harmful outputs, yet challenges remain in defining and controlling hate speech and harmful content 3. Altman acknowledges the potential for AI to cause disinformation problems and economic shocks without requiring superintelligence 4.

It's a certainty there are soon going to be a lot of capable, open-sourced LLMs with very few to none, no safety controls on them.

--- Sam Altman

He stresses the importance of regulatory approaches and using AI to detect and prevent these dangers 4.