https://www.youtube.com/watch?v=hfIUstzHs9A

LLMs are natural language processing computer programs that use deep learning and neural networks to generate text and source code. Some notable ones are GPT-3, GPT-4, LaMDA, BLOOM, and LLaMA.

LLMs belong to a category of models called foundation models

• The term foundation models was coined by a team from stanford when they observed this:
  • AI models were trained on very task specific data to achieve task specific intelligence
  • They found out that we could be moving to a new paradigm where there could be one Foundational model that could drive all the use case models
• These foundational models could drive any other use cases also
• These models are trained on huge amounts of unsupervised data
• Because these models predict based of what they had seen previously these type of models come under the class of generative models
• These models can be used for non generative tasks by introducing small amount of data
  • These can be used for tasks like Named Entity Recognition, or classification
  • This process is called tuning
• If you have very little or un structured data in a process called prompting or prompt engineering to perfrom non generative tasks

Advantages of LLMs

• Performance
• Productivity gains

Disadvantages of LLMs

• Compute
• Trust

Foundation models not just limited to language but are being made for various domains