Meta’s Llama 2 enters the race, challenging ChatGPT and Google’s Bard.
Meta has introduced Llama 2, the latest iteration of their open-source large language model (LLM), building on the success of its predecessor, Llama. Developed in collaboration with Microsoft, Llama 2 is now accessible via various platforms such as Microsoft Azure, Amazon Web Services (AWS) and AI company Hugging Face. The best part is that it is free of charge for research and commercial purposes.
Llama 2 has three pre-trained and fine-tuned models, boasting 7 billion, 13 billion and an impressive 70 billion parameters. In the realm of LLMs, parameters as the repository of knowledge for the model after its training. A higher parameter count generally leads to more precise predictions, allowing the model to leverage a wider scope of contextual information.
Boasting a 40% increase in training data and a context length that’s double its predecessor, Llama 2 significantly outshines Llama 1. Meta claims that Llama 2 surpasses other open-source language models across a range of external benchmarks, demonstrating exceptional performance in reasoning, coding proficiency and knowledge tests.
Meta’s open model shakes up the competition
LLMs are the foundation for chatbots like OpenAI’s ChatGPT and Google’s Bard. For instance, ChatGPT, an acronym for chatbot generative pre-trained transformer, is built upon the GPT large language model. It processes natural language inputs, predicts the subsequent word by analyzing previous inputs and continues until a complete response is generated.
Meta’s LLMs are open-source, meaning that the tech giant intends to publicly share detailed information regarding the new model. This distinguishes it from competitors like OpenAI, which keeps the data and code behind its latest model, GPT-4, inaccessible to outside entities, effectively operating as a black box.
This open-model strategy helps Meta keep pace with its competitors while fostering collaboration and transparency within the AI community. The tech titan strongly advocates for open access to contemporary AI models, asserting that it empowers a new wave of developers and researchers to jointly scrutinize and test these models. Fostering a collaborative community can expedite the identification and resolution of issues and challenges.
Moreover, by studying how others utilize these tools, Meta’s teams can glean valuable insights to enhance the tools and address any potential vulnerabilities that may emerge.
The stage is set for an intriguing face-off as Meta’s Llama 2, the latest contender in the LLM race, gears up to challenge the reigning supremacy of ChatGPT and Google’s Bard. With such intense competition between these advanced language models, AI-driven chatbots will witness exciting and groundbreaking developments.
- Top Generative AI Startups to Watch in 2023
- Databricks Acquires Generative AI Startup MosaicML for US$1.3 Billion
- Google Bard vs. ChatGPT: Which Is the Better AI?
Header image courtesy of Pexels
Press release link: https://about.fb.com/news/2023/07/llama-2/