@theaichampionforbusiness: 𝗪𝗵𝗮𝘁 "𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲" 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗺𝗲𝗮𝗻𝘀 𝗶𝗻 𝗔𝗜, 𝗶𝗻 𝗽𝗹𝗮𝗶𝗻 𝗘𝗻𝗴𝗹𝗶𝘀𝗵 (2026) Inference is what happens when you use an AI model. You ask, it answers. That's it. Most people nod along in AI meetings while the vocabulary goes over their heads, so here's one word decoded properly. 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝘃𝘀 𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 • Training is where the model learns, reading enormous amounts of data and adjusting itself over weeks or months. It happens rarely, in a data centre, long before you touch it. • Inference is where the model works, producing an answer from what it already knows. It happens every single time you press enter. • Training is like the years someone spent studying. Inference is them answering your question on the spot. • The model isn't learning from you when you prompt it. It's applying what it learned already. 𝗪𝗵𝘆 𝗶𝘁 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝘁𝗼 𝘆𝗼𝘂𝗿 𝗯𝘂𝗱𝗴𝗲𝘁 Training costs belong to OpenAI, Anthropic and Google. Inference costs belong to you. Gartner forecasts $42bn of global spending on AI infrastructure in 2026, with $23.3bn of it going on inference against $19bn on training. That's the first year inference has taken the bigger share. Agentic tools make it steeper again, because an agent runs several inference steps per task rather than one, which Gartner puts at around 30 times the cost per task of a simple chatbot. 𝗧𝗵𝗿𝗲𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀 𝘁𝗼 𝗮𝘀𝗸 𝗮 𝘃𝗲𝗻𝗱𝗼𝗿 • Are we paying per user, per token, or per task, and what happens when usage doubles? • How much of the cost comes from agents repeating work rather than people asking questions? • Which jobs genuinely need the biggest model, and which can run on a smaller, cheaper one? Next time someone says inference, hear "using it", and then ask who's paying for it. Which bit of AI jargon should I decode next? #AI #AILiteracy #ArtificialIntelligence #BusinessLeadership #AIAdoption

1104

There are no more comments for this video.

To see more videos from user @theaichampionforbusiness, please go to the Tikwm homepage.