Last Updated on
Llama 2 is the latest update to the newly open-sourced AI from Meta. In fact, the name itself is an abbreviation of Large Language Model Meta Artifical Intelligence – and I do appreciate whoever decided to go with LLaMA over LLMMAI. But what’s in a name? What is Llama 2, what does it do, and why do people seem to care about it so much?
What is Llama 2?
Llama 2 is an AI. An artificial intelligence model to be specific, and a variety called a Large Language Model to be exact. Make sure to check “What is ChatGPT – and what is it used for?” as well as “Bard AI vs ChatGPT: what are the differences” for further advice on this topic. It takes an input of text, written in natural human language, and produces an output of the same kind. It does this by looking at many (many) examples of pre-written text, collectively called the dataset. This dataset is stored as tokens.
Essential AI Tools
Winston AI detector
Originality AI detector
Tokens are very short chunks of text, only a few characters long, a single syllable perhaps. Each token will relate to multiple other tokens, again like syllables form words, and this association is called a parameter.
Think of parameters like neural connections between the synapses in your brain – after all, your brain and Llama 2 are both neural networks, the only difference being that the latter is artificial. These connections, and the frequency with which they are used, allows an LLM to understand how syllables and words and phrases relate to each other, and which are more important than others. It allows the AI to predict the most likely or appropriate next letter in any sequence. By doing so, it can write unique text not necessarily included in its training data. The more parameters an LLM has, the more powerful it is considered.
How powerful is Llama 2?
There are more measures of an AI’s proficiency in knowledge tests and inference in critical thinking than sheer number of parameters. Number of tokens and context length are important in other ways. In order to find what truly makes a difference, there are many third party external benchmarks that AI companies use to size each other up. Compare Llama 2’s 70B parameters (and 2 trillion tokens!) to it’s competition and an interesting picture begins to form.
- Falcon (Technology Innovation Institute of UAE) – 40B
- Llama 2 (Meta) – 70B
- Bard (Google) – 137B
- Megatron-Turing Natural Language Generation (NVIDIA) – 530B
- GPT-4 (OpenAI) – 1.76T
70 Billion of anything sounds like a lot. But in the world of AI, enough is never enough.
Where can you access Llama 2?
Llama 2 is available to download at request from Meta AI. It’s also accessible through hugging face – a hub of sorts for the AI community, providing a user interface to many of the leading AI models. Hugging face hosts many different versions of Llama, and all of them could be considered chatbots (with one even named Llama 2-chat).
Why should you care?
While it is objectively not the most powerful AI of it’s kind, it is special in one respect. Meta AI, in partnership with Microsoft and their Azure Platform, have recently announced that the new technology will be both free for commercial use and open source. This collaboration enables an ecosystem within with developers can build software that relies on Llama 2 and solve valuable problems that would otherwise be impossible.
Open source language models are critically important to safe, fast, and productive development in this field. Nothing short of a billion-dollar-corporation has the funds to train and sustain a comparable AI model. The AI development process can cost millions of dollars per day. This open innovation approach allows the world as a whole to make the most of this costly ground work, building new apps and websites and user interfaces that call on the underlying LLM to perform a specific task. It also allows scrutiny of the underlying codebase and of the bias inherent in the dataset. Open source drives innovation, and this move from Meta CEO Mark Zuckerberg is a positive one for the landscape of the LLM Market.
Is Llama like ChatGPT?
Llama (or LLaMA) is more like GPT-4 than ChatGPT, in that the former is a model, and the latter is an interface. ChatGPT uses GPT-4 but they are not synonymous.