UAE center unveils Falcon LLM, 'better, faster' ChatGPT rival
The model claims to use only 75 percent of GPT-3’s training compute time, and 40 and 80 percent of DeepMind’s Chinchilla NLP and Google’s PaLM-62B compute times, respectively.
DUBAI — Abu Dhabi’s Technology Innovation Institute (TII) this week launched a rival to the generative artificial intelligence (AI) platform that powers the likes of ChatGPT3, claiming their foundational large language model (LLM), named Falcon LLM, performs better.
Falcon LLM, like Open AI’s popular ChatGPT3, learns language by training on large amounts of digital text data to generate advanced conversational abilities that allow human-like exchanges and can interpret nuance to the level of writing speeches and poetry, answering questions, and giving advice.
To do so, Falcon LLM said it uses 40 billion parameters, or parts of the model learned from historical training data, compared to ChatGPT3’s 175 billion, to teach its deep learning tool Generative Pre-trained Transformer 3 (GPT-3), which it claims can outperform the language processing power of existing models.
Falcon LLM claims to use only 75 percent of GPT-3’s training compute time, and 40 and 80 percent of DeepMind’s Chinchilla NLP and Google’s PaLM-62B compute times, respectively.
