Key ideas from Greg Brockman

• Greg Brockman built the first reverse Turing test game in 2008, where humans compete to identify each other while talking to AI bots
• Computing Machinery and Intelligence by Alan Turing in 1950 predicted exactly how modern AI would work through unsupervised learning and reinforcement learning
• The 2012 ImageNet Classification with Deep Convolutional Neural Networks (AlexNet) paper launched the deep learning revolution with a massive performance jump
• OpenAI started in Brockman's San Francisco living room in 2016 with the three-step plan: solve reinforcement learning, solve unsupervised learning, gradually learn more complicated things
• By March 31st, OpenAI aims to make agents the default tool for all employees rather than text editors or terminals
• Pre-training costs millions on month-long GPU runs while post-training takes just days to teach models how to behave
• OpenAI's unreleased AI system proved a quantum physics hypothesis wrong, marking the first time it felt like the system was truly thinking
• Individual OpenAI employees now want 1,000 dedicated GPUs each, but only 10 million GPUs exist globally