My notes from Tom Bilyeu's 2h45m interview of Emad Mostaque (Founder and CEO of Stability AI)
🟢 In March 2023, he signed an open letter calling for "all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4".
🟢 Stability AI is one of the TIME100 Most Influential Companies of 2023.
🟢 People misinterpret ChatGPT and other AI as "programs". Instead, they should be thought of more as "filters" or codecs.
🟢 Some of these filters are only 100 GB or even 2 GB. Imagine downloading an AI onto a tablet for use in Africa even without an internet connection!
🟢 Now that Nvidia and others have "cracked" how to scale the computation side, companies don't have that constraint. They can invest more money into building more powerful AI models.
🟢 What we've seen so far is the result of only tens of millions of dollars of investment per model. Companies are ready to invest $10B+. Imagine how much more powerful the tools will soon be.
🟢 Even the people most accustomed to exponential growth curves and intimately aware of AI (such as the "godfather of AI") have been shocked recently at the pace and acceleration of progress.
🟢 This leads experts to admit that even trying to predict what the world will be like in 1-3 years is too difficult.
🟢 There is a very significant chance that the pursuit of creating a super AI (Artificial General Intelligence—AGI) will lead to an extinction or enslavement of humanity.
🟢 Raising kids nowadays with so much uncertainty is scary.
🟢 Mostaque disagrees with Ray Dalio's concern about civil war with physical conflict. Both agree that the tensions are high and rising. However, Mostaque believes that governments will be able to distract enough would-be soldiers with AI girlfriend sexbots, essentially.
🟢 There is going to be a massive mental health crisis soon. Children born nowadays might feel like it's impossible to feel unique or add value. Older people might feel unwilling to put in the effort to "keep up". How are we collectively going to address the existential crises that so many people will be having?
🟢 The 2024 and 2028 American elections are going to be disasters. There will be more misinformation (more convincing deepfakes) than ever before. Even if people recognize them as fakes, the damage will be done, because fewer people will vote, and those who do could be subconsciously influenced.
🟢 In 1-3 years, the entertainment industry might be completely replaced by one where you can ask your own AI to create entire feature-length Hollywood-quality films for you. The scripts will be written by AI, the actors will be photorealistic and can look or sound like any real or imagined person or creature.
🟢 So far, the recent approach to AI (such as ChatGPT) has been to scrape the whole internet and be a generalist. It seems that an even better approach is to have an AI manager of other specialist AIs. So you could make your request, and the manager AI will outsource to dozens or hundreds or thousands of expert AIs that were specially designed for certain skills.
🟢 Really talented humans can be many times more productive with these tools. Scientists might cure cancer, solve climate change, invent cold fusion reactors, etc all within a number of years EVEN WITHOUT any improvement to AI's reasoning ability.
🟢 A surprisingly small number of people are working in the field of AI, and even fewer people are working in the field in a way that cares about preventing the doomsday scenarios. So you individually could make an impact if you make it your mission.
It was an amazing conversation (published 2023-07-18).
I've been loving Tom Bilyeu's long interviews lately.