Stability AI Founder: Halting AI Training is Hyper-Critical For Our Security w/ Emad Mostaque |EP#36
The podcast discusses the rapid advancement of AI, particularly large language models, and the concerns around their development speed, potential regulation, and the future of open vs. closed AI approaches. Emad Mostaque, CEO of Stability AI, shares insights on the complexity and uniqueness of these models, and the potential of large language models in achieving AGI. The podcast discusses the rapid advancement of large language models, their potential risks, ethical concerns, and the impact on the job market. The podcast discusses the rapid advancements in AI computing, Emad Mostaque's personal health practices, the debate on slowing down AI development, and the potential future of AI in augmenting human capabilities. The conversation revolves around the ethical and business implications of open AI, its potential impact on society, particularly in education, and the argument against arti
“they don't have the chips they don't have the KN how we've seen no evidence of that and they don't want to overthrow their system they want to perpetuate it and you know what the best way for them to get an AI is is to send someone with a flash drive into Deep Mind or open Ai and just download it”
Your call:
openEmad Mostaque: There should be a public debate around the regulation of AI technology, as it is becoming increasingly powerful and transformative.
“this is the time for a public debate around it and you know everyone should realize that this is taking off crazy um and that's what democracy is about”
Your call:
openEmad Mostaque: Every child in the world will have access to their own AI teacher in five years.
“we're going to see I think the estimate that Ray Chiswell talked about and we've talked about at singular University for a while is we're going to see a century worth of progress in the next 10 years”
Your call:
Numbers that matter
50%Half of all code on GitHub now is AI generated.
640The UK 640 chips equivalent of 30 to 100,000 chip supercomputers using the new generation technology.
6 monthsTime until the new generation of Nvidia GPUs and TPUs will be available.
fewExpected number of large language models in the future.
40%Large language models have made programmers 40% more efficient according to a study.
3GPT-4 can pass the level three Google programmer exam out of the box without tuning.
2GPT-4 can pass the medical licensing exam, which the speaker never finished doing.
20-40The new supercomputers are 20-40 times faster than the current fastest supercomputer.
20-40 times fasterNew supercomputers are expected to be 20-40 times faster than current ones.
76%76% of children in refugee camps achieved literacy and numeracy in 13 months with the software.
25 wattsThe human brain consumes 25 watts, compared to 400 Watts for running AI on two H100s, highlighting the efficiency of AI.
1.5 millionpeople still use AOL, indicating the slow integration of new technologies into existing systems.