My hilarious CTO friend texted me last Saturday and shared some news…
The same Mark (Meta), who has been accused of mishandling personal data, launched its new language model last Friday. It is a significant milestone for Meta, scientists, researchers, the AI community, and I dare to say… society.
After diving into their official paper and model card for three nights in a row with a couple of highly credible yet geeky friends (CTO of an AI startup and a lead AI researcher), it seems like we are on the verge of a revolution. This can be the model that will shape how AI works in the future.
In a nutshell, Meta has developed an OPEN-SOURCE (which is not Facebook-like) AI language “generator” to democratize access to sophisticated language models so people can use it for research. This democratization could lead to more innovation and creativity while addressing ethical and social concerns in AI.
While some (including myself) may have concerns about Meta’s track record with data privacy issues and misinformation, it’s important to note that LLahMA is being released as an open-source resource. This means that the vast amount of data collected by Meta can be used to train AI models that are ethical and beneficial for society.
Training smaller models (yes, not BIG, as my friend mentioned) like this can open up huge possibilities for creative text generation and problem-solving. Through 1.4 trillion tokens across 20 languages and Latin/Cyrillic alphabets, LLaMA helps researchers push past boundaries in ways never before possible – from predicting protein structures (to potentially developing new medicine) to answering reading comprehension questions – without exhausting all your resources. And because access remains limited due to required computations, they’re granting it on an individual basis so everyone gets their chance at playing genius scientist. That’s pretty amazing!
