How to control AI most effectively? – An interview with GPT-3 AI Agent, which ponders on its future, when it may itself actually control its own makers)
20/1/2021
Tony Czarnecki, Managing Partner, Sustensis
Introduction:
This interview has been carried out with Generative Pre-trained Transformer 3 (GPT-3), now also known as INSTRUCT, is an Artificial Intelligence agent, one of the most advanced Natural Language Processors (NLP) supported by 175 billion tokens (words, split into 3-4 characters bits), created by Open-AI in July 2020. Its database includes all Wikipedia and millions of articles and reliable sources of information from all knowledge domains. To use it, one must adjust the GPT-3 to particular needs. That includes the creation of a few sentences’ context PROMPT (a statement or a question), to which GPT-3 writes a COMPLETION. Others include setting several parameters, such as TEMPERATURE (probability of selecting a given token), which may affect the quality of the answers.
I have tried to convince the AI Agent that we need to control AI with the help of the first Transhumans, an idea that I have developed in a paper for 2022 AI conference in Seattle.
Read the whole interview here.