3 January 2025

Things LLMs Cannot Do

  • Lack of True Understanding:
    • They operate on probabilities and patterns based on the training data
    • They have no understanding on the meaning of language
    • They have no grasp of concepts on the basis of which they generate grammatically correct sentences

  • Hallucination and Bias:
    • They have the propensity to generate incorrect and nonsensical information
    • They mimic the biases in their training data that leads to unfair or discriminatory output

  • Sensitivity to Input:
    • Minor changes in input can drastically alter output that makes it difficult to control response

  • Lack of Common Sense and Real-World Knowledge:
    • They lack true common sense and worldly experience, the most they can access and process is the information based on their training data
    • They fundamentally lack the understanding of everyday situations and interactions

  • Ethical Concerns:
    • They can be used to generate misleading and even harmful content
    • They can be misused that have ethical implications

  • Limited Creativity and Originality:
    • They lack basic sense of creativity that is outside the fold of the patterns in training data to generate truly original concepts and ideas
  • Emotions and Consciousness:
    • They lack feelings, self-awareness and have no way of experiencing emotions
  • Reliance on Data:
    • They are only as good as the data they have been trained on and are a reflection of it
  • Physical Actions:
    • They are software programs in the digital realm and cannot interact with physical world