ChatGPT offered the highest level that artificial intelligence supported chatbots can reach today to the end user. Language models that rivaled the language model that fed ChatGPT had also become an area where other technology companies concentrated.
On the one hand, Google continues to develop its own language model, LaMDA, and there was great news from Meta last week. Meta introduced the language model LLaMA, which aims to contribute to the scientific world. The language model, open only to scientific studies, was leaked in its first week.
LLaMA is already available to anyone who wants to:
4chan user ‘llamanon’ leaked versions of LLaMA with 7 billion and 65 billion parameters via torrent. The non-public language model was made available through a library published on GitHub.
A short time later, a library on GitHub for the acquisition and use of all versions of LLaMA was launched by a user named ‘shawwn’. All versions of LLaMA have a total size of 219 GB.
Of course, in order to run the language model efficiently, you need to have graphics card hardware with extremely high processing power and therefore a very salty price tag.
Meta has yet to take action to remove the leak from the internet.
Can LLaMA rival OpenAI’s language models?
Some users who performed tests by downloading leaked versions of LLaMA stated that the language model is at least as advanced as the ‘davinci’ model. Until last week, OpenAI’s davinci model was a language model that anyone could access and create their own chatbots.
We don’t know yet how advanced the language model is. But it seems OpenAI continues to rank at the top, with the ‘gpt-3.5-turbo’ language model that makes up today’s ChatGPT.
Why hasn’t Meta made this language model publicly available?
ChatGPT’s responses were also a sign that the language model could be abused (or even begun to) in some way. Since Meta is also a huge company and doesn’t want to tarnish its name, it actually didn’t want to indirectly support this abuse.
For this reason, the company has made the language model available only to scientists working in the field of artificial intelligence. The purpose of the language model was to be a basis for enabling the development of more advanced language models in the more advanced future.
8774
