Meta has Become Clumsy and its AI Galactica is the Proof
Meta AI was very hopeful about their AI Galactica and it is not what they hoped for
Galactica is an artificial intelligence developed by Meta AI (formerly known as Facebook Artificial Intelligence Research) with the intention of using machine learning to “organize science.” It’s caused a bit of a stir since a demo version was released online last week, with critics suggesting it produced pseudoscience, was overhyped, and not ready for public use.
The tool is pitched as a kind of evolution of the search engine but specifically for scientific literature. Upon Galactica’s launch, the Meta AI team said it can summarize areas of research, solve math problems and write scientific code.
Meta’s new scientific AI system turns out to generate well-written research papers on the benefits of committing suicide, practicing antisemitism, and eating crushed glass.
Like Rodin’s The Thinker, there was plenty of thinking and pondering about the large language model (LLM) landscape last week. There were Meta’s missteps over its Galactica LLM public demo and Stanford CRFM’s debut of its HELM benchmark, which followed weeks of tantalizing rumors about the possible release of OpenAI’s GPT-4 sometime over the next few months.
The online chatter ramped up last Tuesday. That’s when Meta AI and Papers With Code announced a new open-source LLM called Galactica, that it described in a paper published on Arxiv as “a large language model for science” meant to help scientists with “information overload.”
The “explosive growth in scientific literature and data,” the paper’s authors wrote, “has made it ever harder to discover useful insights in a large mass of information.” Galactica, it said, can “store, combine and reason about scientific knowledge.”
Galactica immediately garnered glowing reviews: “Haven’t been so excited by a text LM for a long time! And it’s all open! A true gift to science,” tweeted Linxi “Jim” Fan, a Nvidia AI research scientist, who added that the fact that Galactica was trained on scientific texts like academic papers meant that it was “mostly immune” from the “data plagues” of models like GPT-3, which was trained on texts trained on the internet at large.