Generative AI is evolving at a fast pace.
Llama 2 is a generative AI model but there are a few key differences between Meta’s Llama 2 and its competitors—including OpenAI’s GPT-4 (ChatGPT) and Google PaLM 2 (Bard). Check the differences here.
2023 is certainly shaping up to be the year of AI, with companies racing to launch their own generative AI models. Meta has followed suit with an upgraded version of its LLaMa LLM model, called Llama 2. Meta has released the model as open source—meaning that it is available for free for both commercial and research purposes.
Meta and Microsoft have partnered to distribute Llama 2—which is now available in the Azure AI model catalog. This will allow developers using Microsoft Azure to use and benefit from Llama 2.
Now, while Llama 2 is a generative AI model, or more specifically, a Large Language Model (LLM), there are a few key differences between Meta’s Llama 2 and its competitors—including OpenAI’s GPT-4 (ChatGPT) and Google PaLM 2 (Bard).
Key Differences Between Meta Llama 2, GPT-4 and PaLM 2:
Llama 2 Is Less Powerful Than GPT-4 and PaLM 2, Meta Admits
Percy Liang, director of Stanford’s Center for Research on Foundation Models, points out that Meta’s research paper acknowledges that Llama 2 is less powerful than GPT-4 and that there is a performance gap between the two models. In terms of benchmarks, Llama 2 model also performs slightly below its rivals—GPT-4 and PaLM 2.
Trained Using Fewer ‘Tokens’ Compared to Competition
For those who are not familiar with the term, ‘tokens’ in this context refer to the text that a generative AI model was trained on. For example, ‘bat’ and ‘cool’ are both tokens. The more tokens an AI model is trained on, the better it will be. According to TechCrunch, Llama 2 was trained on two million tokens, which is twice as many as the 1.4 trillion tokens that the original Llama was trained on. For comparison, Google’s PaLM 2, its immediate rival, was trained on 3.6 million tokens, according to a CNBC report from May.
Language Support
Meta’s Llama 2 supports fewer languages than Google’s PaLM 2 and OpenAI GPT-4, with 20 languages compared to PaLM2’s 100, and GPT-4’s 26. Here, Google seems to be ahead. Google Bard, which used the PaLM 2 model now even supports nine Indian languages: Hindi, Tamil, Telugu, Bengali, Kannada, Malayalam, Marathi, Gujarati, and Urdu. And, overall, Google is now supporting over 40 languages for Bard including Bahasa Indonesia.
