flâneur

Generating Fake Conversations by fine-tunning OpenAI’s GPT-2 on data from Facebook Messenger | Svilen Todorov

svilentodorov.xyz · 2,157 words · saved by 1 readers

One of the most interesting problems in NLP has always been human-like conversation and many are still considering passing the Turing Test as the holy grail of the field. In this post, I show how to use a state of the art model on your own data (I use my own messages sent on facebook) to generate (somewhat) realistic conversations.

One of the most interesting problems in NLP has always been human-like conversation and many are still considering passing the Turing Test as the holy grail of the field. In this post, I show how to use a state of the art model on your own data (I use my own messages sent on facebook) to generate (somewhat) realistic conversations. In February 2019 OpenAI released information on their new state of the art language model which created a lot of buzz within the community. While few disagree the results they included are better than anything we’ve seen before (even if mainly because they made a bi

related reading