HPR3379: Linux Inlaws S01E34: The one with the intelligence




Hacker Public Radio show

Summary: In this fourth part of our three-part miniseries on Deep and Machine Learning our two heroes shed some light on a DL architecture called Generative Pre-Trained Transformer (GPT), a pretty sophistic piece of software that fools most humans when it comes to authoring text (ideal for budding writers with a block in place). Other topics of discussion includes OpenAI (the company behind this framework), Elon Musk, Bitcoin, Microsoft and if the GPT can actually pass the Turing test. All will be revealed - don't miss this episode! Links: OpenAI: https://openai.com GPT: https://openai.com/projects The Turing Test: https://en.wikipedia.org/wiki/Turing_test GPT-2 source code: https://github.com/openai/gpt-2 GPT meta-progamming: https://www.lesswrong.com/posts/zZLe74DvypRAf7DEQ/meta-programming-gpt-a-route-to-superintelligence GPT-3 interview: https://www.youtube.com/watch?v=PqbB07n_uQ4 DSDS: https://en.wikipedia.org/wiki/Deutschland_sucht_den_Superstar GPT-3 sample 1: https://linuxinlaws.eu/files/padawans.txt GPT-3 sample 2: https://linuxinlaws.eu/files/HGttG.txt