Sam Altman interview Lex Fridman

Sam Altman interview Lex Fridman
Sam Altman OpenAI interview Lex Fridman

Finally! In a recent interview with Lex Fridman, Sam Altman, CEO of OpenAI, talks about the possibilities and dangers of AI in this current moment in the history of human civilization. He discusses the development of GPT-4, the language model trained on a large dataset that contains human knowledge and wisdom.

Topics Discussed

  1. The development of GPT-4 and its capabilities
  2. The selection of the pre-training dataset for GPT-4
  3. The science of human guidance in AI
  4. The importance of usability, alignment, and ethics in AI
  5. The possibility of AI possessing wisdom
  6. The struggle of GPT-4 with certain tasks, such as counting characters
  7. OpenAI's iterative process of building in public and receiving feedback from users
  8. The bias of GPT-3 and the need for personalized control over AI
  9. The challenge of understanding the reasoning capability of AI
  10. The potential societal transformation brought about by AI

GPT-4 by OpenAI

Altman believes that we are on the verge of a fundamental societal transformation, where AI systems will surpass the collective intelligence of the human species by many orders of magnitude. While this is exciting due to the innumerable applications that AI can empower humans with, it is also terrifying because of the power that superintelligent AGI wields to intentionally or unintentionally destroy human civilization. Altman explains that GPT-4 is a system that we will look back at and say it was a very early AI, slow, buggy, and not very useful. However, it points to something that will be really important in our lives, even though it will take a few decades to evolve.

Understanding GPT-4

GPT-4 is a language model that has been trained on a large pre-training dataset to create background wisdom knowledge that is contained within the internet. The system has the ability to do some kind of reasoning, which is remarkable. Altman explains that the science of human guidance is a crucial aspect of AI, as it helps align the model to what humans want it to do.

The Pre-Training Dataset

The pre-training dataset of GPT-4 is a huge amount of effort that is pulled together from many different sources, including open-source databases of information, partnerships, and things on the internet. The dataset includes a lot of content in the world, more than most people think. However, Altman believes that too much processing power is going into using the models of the database instead of using the model as a reasoning engine.

Creating GPT-4

To create GPT-4, Altman explains that many pieces have to come together, and existing ideas have to be executed well at every stage of the pipeline. The evaluation of the model is essential to measure how good it is, and the most critical aspect is how much utility it provides to people. Altman believes that understanding how much value and utility a model provides to people is something that is understood better now than before.

Building in Public

Altman believes that building in public is essential, and OpenAI puts out technology because they believe it is important for the world to get access to it early to shape the way it's developed, to help find the good things and the bad things, and provide feedback. The iterative process of putting things out, finding the great parts, the bad parts, improving them quickly, and giving people time to feel the technology and shape it with us and provide feedback is crucial.

Conclusion

In conclusion, GPT-4 is a significant advancement in AI technology, and while it has its limitations, it points to something that will be really important in our lives in the future. However, the dangers of AI must also be considered, as superintelligent AGI can have the power to destroy human civilization intentionally or unintentionally. The science of human guidance is crucial to align AI to what humans want it to do, and the evaluation of the model is essential to measure how much utility it provides to people. Building in public is essential to shape the development of AI and to get feedback from users to improve it quickly.