Meta’s new LIMA language model reaches GPT-4 level
Meta’s AI experts introduce LIMA, a new language-model that achieves GPT-4 or Bard levels of performance in test scenarios.
LIMA stands for Less is More for Alignment, and its name gives a hint at the model’s purpose: it is meant to demonstrate that with a pre-trained AI system, just a few examples will suffice to produce high-quality results.
Meta has manually selected 1,000 different prompts from sources like WikiHow and Reddit, as well as other research papers.
Source:
https://the-decoder.com/metas-new-lima-language-model-reaches-gpt-4-level/
