During a recent interview with Lex Fridman, CEO of OpenAI Sam Altman discussed several topics, including whether the language model GPT has an inherent bias. Altman admitted that the term “woke” has changed over time and that it is prejudiced, according to the interview published on Saturday. He also mentioned that OpenAI made significant improvements to the GPT model between versions GPT-3.5 and GPT-4, and expressed appreciation for critics who recognize the progress made by the company.
In addition, Altman addressed Elon Musk’s criticism of OpenAI’s AGI safety research, expressing empathy for Musk’s concerns while also encouraging him to focus on the hard work of addressing AI safety issues. Altman also shared his thoughts on AI risks, stating that many of the predictions regarding AI’s capabilities and safety issues have been inaccurate.
Altman also discussed the topic of “jailbreaking,” which refers to users modifying AI models outside of the intended bounds. While Altman emphasized the value of giving users control over the models, he acknowledged that the existence of jailbreaking shows that the problem has not yet been fully solved.
The conversation also touched on more general subjects, such as the meaning of life, the difference between fact and fiction, and parenting tips. Throughout the interview, Altman emphasized the importance of intellectual honesty, recognizing mistakes, and learning from them.
