SIGNAL//DESK
AI/MLsrc: Roost FA glossary v2

Latency

How long the model takes to reply — typically 1-30 seconds depending on length. High latency can disrupt real-time client interactions if the FA is waiting for a response during a meeting.

Latency refers to the response time of an AI model, typically ranging from 1 to 30 seconds based on output length. In professional settings, high latency is a critical performance metric because it can cause significant friction during real-time client interactions.

Latency is the temporal delay between the submission of a prompt and the completion of the generated output, measured in seconds as a function of computational load and token sequence length. In operational contexts, this duration—typically 1-30 seconds—must be minimized to prevent latency-induced disruption during synchronous client-facing interactions.

evolution

  1. 1956 · history
    Dartmouth Workshop

    The concept of computational response time was first implicitly addressed as a bottleneck for real-time AI problem solving.

  2. 2012 · history
    AlexNet Breakthrough

    The rise of deep learning on GPUs shifted the focus of latency from algorithmic complexity to hardware-accelerated inference speed.

  3. 2022 · history
    LLM Mainstream Adoption

    The release of ChatGPT highlighted 'Time to First Token' (TTFT) as a critical metric for user experience in conversational AI.

  4. 2023 · history
    Speculative Decoding

    New inference optimization techniques emerged to reduce latency by using smaller models to predict tokens for larger models.


← all terms