Skip to main content

Large Language Models

Data used to train LLMs

  • LLMs are trained in an unsupervised manner on open source, licensed and borrowed data
  • Responses are refined using Q/A pairs
  • Reinforcement learning with human feedback is used to reward LLMs to give appropriate responses

LLMs can (seem to) be creative

  • Context-based learning + randomness allow the LLMs to generate surprising outputs
  • Can use LLMs to:
    • Identify weakly similar concepts from different disciplines and help understand different disciplines
    • Generate narratives etc

Challenges in LLMs

  • LLMs may seem to "lie" and "hallucinate"
  • This is some function of data, learning, search and probability
  • Remember that the output of a LLM is determined by both what the system has been trained on and what information you give it
  • Prompt engineering means tailoring your questions and input so you can get the most out of an LLM

Prompting Techniques

  • Prompting techniques are about formatting your input to provide a context that the LLM ca use more effectively, restricting the probability distribution of the words it chooses to sample next

Training phases of LLM

....