(2019-02-17) Azhar Gpt2

Azeem Azhar on OpenAI and GPT2.

OpenAI, a private research group, demonstrated GPT2, an AI text generator, which the group reckoned was too dangerous to be released publicly.

I have some misgivings about the way hyperbolic manner in which GPT2 was announced. Nvidia’s Anima Anandkumar says it better than me: “What [OpenAI is] doing is opposite of open. It is unfortunate that you hype up + propagate fear + thwart reproducibility + scientific endeavor.”

While it doesn’t hurt to be have some methods or policies of review, reflection and attentuation in place, the question is less about preventing the diffusion. It is about how we start making the systemic changes to build resilience into our systems. And systemic means that. Not just technical solutions, like blockchain-based approaches, that could ensure provenance and authority, but also investments in building social capital and trust across societies that help them withstand disinformation. It also stretches to investments in improving the media literacy and analytical toolkits of citizens. For example, globally about two-thirds of people do not know that the Facebook newsfeed is algorithmically generated.