source: techcrunch ai: so you’ve heard these ai terms and nodded along; let’s fix that
level: technical
artificial intelligence has spawned a dense vocabulary that can leave even tech-savvy people feeling lost. terms like agi, llm, and chain-of-thought appear constantly in news and product descriptions. this glossary defines the most common ones in simple language. it covers foundational ideas such as neural networks and deep learning, which use layered structures inspired by the brain to find patterns in data. it also explains practical concepts like fine-tuning, where a model gets extra training on specialized data to improve performance for a specific task.
many terms describe how ai systems operate. inference is the act of running a trained model to get predictions, while training is the earlier phase where the model learns from data. compute refers to the processing power—often from gpus—that makes both possible. parallelization speeds things up by doing many calculations at once. memory caching stores previous results to avoid redundant work. distillation creates a smaller, faster model by learning from a larger one. diffusion models generate images or audio by learning to reverse a noising process, and gans pit two networks against each other to produce realistic outputs.
other terms address reliability and autonomy. hallucination means the model makes up false information, a major quality problem. chain-of-thought reasoning breaks problems into steps for better answers, especially in logic or coding. ai agents are tools that perform multistep tasks on a user's behalf, like booking tickets or writing code. coding agents are a specialized type that can write, test, and debug software autonomously. api endpoints are the hidden buttons that let agents interact with other services. open source models, like meta's llama, allow public inspection and modification, while closed source models keep their code private.
why it matters: understanding these terms helps data scientists and ai practitioners communicate clearly, choose the right tools, and avoid costly misunderstandings about model capabilities and limitations.
source: techcrunch ai: so you’ve heard these ai terms and nodded along; let’s fix that