[Download full paper -> https://dhanjani.com/docs/Dhanjani_AI_GPT3.pdf]
AI Powered Misinformation and Manipulation at Scale #GPT-3
Risks of autoregressive language models and the future of prompt engineering.
A bunch of very smart people got together and built a bot. They programmed this bot to read the entirety of the Internet. Having read most of the stuff on the Internet, this bot is now pretty great at knowing what word most probably comes next, and the word after that word, if you give it a bunch of words to start with. Like how the iMessage app suggests 3 words you probably will type next.
This autocomplete bot can manipulate people on social media and spew political propaganda, argue about the meaning of life (or lack thereof), disagree with the notion of what constitutes a hot-dog versus a sandwich, take upon the persona of the Buddha or Hitler or a dead family member, write fake news articles that are indistinguishable from human written articles, and also produce computer code on the fly. Well, among other things.
This bot is called GPT-3 and the very smart people are the researchers from OpenAI.
GPT-3 has captured much of mainstream attention, and rightfully so - colorful conversations on Turing completeness and perceived consciousness, even amongst AI Scientists who know the technical mechanics. The chatter on perceived consciousness does have merit - it’s quite probable that the underlying mechanism of our brain is a giant autocomplete bot that has learnt from 3 billion+ years of evolutionary data that bubbles up to our collective selves, and we ultimately give ourselves too much credit for being original authors of our own thoughts (ahem, free will).
In this document, however, I’d like to share my thoughts on GPT-3 in terms of risks and countermeasures, and discuss real examples of how I have interacted with the model to support my learning journey.
Two ideas to set the stage:
- OpenAI is not the only organization to have powerful language models. The compute power and data used by OpenAI to model GPT-n is available, and has been available to other institutions, nation states, and anyone with access to a terminal and a stolen credit-card.
- There exist more powerful models that are unknown to the general public. The ongoing global interest in the power of Machine Learning models by institutions, governments, and focus groups leads to the hypothesis that other entities most likely not only have even more powerful models than GPT-3, but that these models are already in use and have been in use for some time to support various use-cases. These models will continue to become more powerful in time.
[Download full paper -> https://dhanjani.com/docs/Dhanjani_AI_GPT3.pdf]
