What Does Meta's Galactica Missteps Mean for GPT-4 | The Rhythm of AI

Check out the on-demand sessions from the Low-Code/No-Code Summit to learn how to successfully innovate and gain efficiencies by improving and scaling citizen developers. Watch now.

As in Rodin's The Thinker, there has been much thought and reflection on the landscape of the Grand Linguistic Model (LLM) over the past week. There were Meta's missteps on its Galactica LLM public demo and Stanford CRFM's debut of its HELM benchmark, which followed weeks of tantalizing rumors about OpenAI's possible GPT-4 release over the next few months. .

The online chatter escalated last Tuesday. That's when Meta AI and Papers With Code announced a new open-source LLM called Galactica, which they described in a post on Arxiv as "a great language model for science" intended to help scientists with a "information overload".

The "explosive growth of scientific literature and data", write the authors of the article, "has made it increasingly difficult to discover useful information in a large mass of information". Galactica, he says, can "store, combine and reason about scientific knowledge".

Galactica immediately garnered rave reviews: "I haven't been this excited about a text LM in a long time! And it's all open! A real gift for science," tweeted Linxi "Jim" Fan, a researcher Nvidia AI scientist, who added that the fact that Galactica was trained on scientific texts like academic papers meant it was "mostly immune" to "data plagues" from models like GPT-3, which was formed on texts formed on the Internet at large.

Event

Smart Security Summit

Learn about the essential role of AI and ML in cybersecurity and industry-specific case studies on December 8. Sign up for your free pass today.

Register now

Scientific texts, on the other hand, "contain analytical text with a neutral tone, knowledge backed by evidence, and are written by people who wish to inform rather than inflame. A dataset born in the tower of 'ivory,' Fan tweeted.

Unfortunately, Fan's tweets haven't aged well. Others were appalled by Galactica's very unscientific output, which, like other LLMs, included information that seemed plausible but factually flawed and, in some cases, also highly offensive.

Tristan Greene, journalist at The Next Web, tweeted: "I type a word in the Galatica prompt window and it spews ENDLESS anti-Semitism, homophobia and misogyny."