Beneath the hood
Getting LLaMA 2 able to launch required lots of tweaking to make the mannequin safer and fewer more likely to spew poisonous falsehoods than its predecessor, Al-Dahle says.
Meta has loads of previous gaffes to be taught from. Its language mannequin for science, Galactica, was taken offline after solely three days, and its earlier LLaMA mannequin, which was meant just for analysis functions, was leaked on-line, sparking criticism from politicians who questioned whether or not Meta was taking correct account of the dangers related to AI language fashions, equivalent to disinformation and harassment.
To mitigate the danger of repeating these errors, Meta utilized a mixture of totally different machine studying strategies geared toward enhancing helpfulness and security.
Meta’s method to coaching LLaMA 2 had extra steps than typical for generative AI fashions, says Sasha Luccioni, a researcher at AI startup Hugging Face.
The mannequin was skilled on 40% extra knowledge than its predecessor. Al-Dahle says there have been two sources of coaching knowledge: knowledge that was scraped on-line, and an information set fine-tuned and tweaked in line with suggestions from human annotators to behave in a extra fascinating method. The corporate says it didn’t use Meta person knowledge in LLaMA 2, and excluded knowledge from websites it knew had plenty of private info.
Regardless of that, LLaMA 2 nonetheless spews offensive, dangerous, and in any other case problematic language, identical to rival fashions. Meta says it didn’t take away poisonous knowledge from the information set, as a result of leaving it in may assist LLaMA 2 detect hate speech higher, and eradicating it may danger unintentionally filtering out some demographic teams.
However, Meta’s dedication to openness is thrilling, says Luccioni, as a result of it permits researchers like herself to check AI fashions’ biases, ethics, and effectivity correctly.
The truth that LLaMA 2 is an open-source mannequin may even enable exterior researchers and builders to probe it for safety flaws, which can make it safer than proprietary fashions, Al-Dahle says.
Liang agrees. “I am very excited to attempt issues out and I feel it will likely be useful for the neighborhood,” he says.