Meta discreetly introduces Llama 2 Long AI, outperforming GPT-3.5 Turbo and Claude 2 in certain tasks
September 29 2023
During its annual Meta Connect conference, Meta Platforms unveiled several new AI features for Facebook, Instagram, and WhatsApp. However, the most significant news came from a paper published by Meta researchers, introducing “Llama 2 Long” – a new AI model built on Meta’s open-source Llama 2 model. The new model outperforms some of the top competitors, including OpenAI’s GPT-3.5 Turbo, in generating responses to longer user prompts. The team trained the model using more longer text data sources and applied a modification to the positional encoding crucial for the model to process longer sequences. As a result, Llama 2 Long’s improved performance in several benchmark tasks has been met with enthusiasm from the open-source AI community.
Back to Breaking AI News
What does it mean?
- Meta Connect conference: An annual event where Meta Platforms, also known as Facebook, demonstrates its new technologies and tools.
- AI: Short for Artificial Intelligence, it refers to the simulation of human intelligence processes by machines, especially computer systems. These processes include learning, reasoning, self-correction, etc.
- Open-source: In context of computing, it refers to a type of software whose source code is freely available to the public, allowing anyone to inspect, modify, and distribute the software.
- Llama 2 model: A type of Artificial Intelligence model developed by Meta Platforms. The specifics of the model are not explained in the text.
- OpenAI's GPT-3.5 Turbo: A specific model of artificial intelligence developed by OpenAI. It's designed to generate human-like text based on the prompts it receives.
- User prompts: Directions or instructions users provide to an artificial intelligence system to produce desired output.
- Positional encoding: In the context of AI, it refers to a technique used in model training to help the model understand the position or order of data in a sequence.
- Benchmark tasks: Standardized tasks or tests that are used to measure and compare the performance of different systems or models.
- Open-source AI community: A community of people who work on, use, or have an interest in open-source artificial intelligence technology. They communicate and collaborate to improve the technology and share their findings and advancements freely with others.