AI Models<\/i>: We think models are beginning an age of \u201cintelligence-as-a-service\u201d, or IQaaS, and the winners will be decided in a race for talent, data, and compute power. Scaling model performance also means scaling high quality datasets. We\u2019re seeing models perform as well when trained on pruned and curated datasets 30% of the size, which presents an opportunity for innovative engineering teams. Synthetic data and user feedback are also potential frontiers to explore. <\/i><\/p>\n<\/li>\n Compute power and cloud platforms<\/i>: Hardware accelerators \u2014 predominantly GPUs (graphics processing units) \u2013 are the foundation for AI\u2019s power. The insatiable demand for GPUs has led to ripple effects across our entire economy, from stressing our power grids, to bolstering the growth of the semiconductor supply chain, to re-accelerating demand for cloud services. As AI becomes more mainstream, we expect inference workloads to be drastically more compute intensive. Servicing 100 million GPT-4 users could cost 4x more compute per day than the amount spent on training the model. As we begin to deploy AI in production, there is an opportunity for AI startups to innovate across training and inference stacks and bring down costs dramatically.<\/p>\n<\/li>\n Developer tooling<\/i>: One of the big shifts since the explosion of AI in 2022, has been the democratization of AI and the rise of the AI engineer. A new crop of AI Ops tooling has emerged, which has made it dramatically easier to not only train, fine tune, and deploy models, but build on top of those models to create useful applications. Due to the emergence of a vibrant AI Ops tooling layer, the ~30M developers in the world can access AI, which creates potential for an explosion within the application layer.<\/p>\n<\/li>\n AI applications<\/i>: We\u2019re seeing just the beginnings of AI applications adoption, and we predict that many more use cases will emerge. Consumer and enterprise interest in AI is already gaining momentum from creative domains with tools like Runway<\/a> and Tome<\/a>, to professional use cases like Replit<\/a> and Aurora Solar<\/a>. However, startups building AI applications will need to move quickly and create new behavioral shifts if they want to beat incumbents.<\/p>\n<\/li>\n<\/ul>\n<\/blockquote>\n\n Download (PDF, 13.85MB)<\/a><\/p>\n \ub2e4\uc74c\uc740 Olyber Wyman\uc778 UK Finance\uac00 Oliver Wyman\uacfc \ud611\ub825\ud558\uc5ec \ubc1c\ud589\ud55c The impact of AI in financial services: opportunities, risks and policy considerations<\/a>\uc785\ub2c8\ub2e4. \uc81c\ubaa9\ucc98\ub7fc \ud589\uc815\ubd80\uc640 \ud611\uc5c5\ud55c \uacb0\uacfc\ubb3c\uc774\uae30 \ub54c\ubb38\uc5d0 \uc0ac\ub840 \ubc0f \uc815\ucc45\uc801\uc778 \uc870\uc5b8\uc744 \ub2f4\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n![\"\"](\"https:\/\/i0.wp.com\/www.smallake.kr\/wp-content\/uploads\/2023\/12\/revolution1.png?resize=1130%2C637&ssl=1\")
<\/a>
\n![\"\"](\"https:\/\/i0.wp.com\/www.smallake.kr\/wp-content\/uploads\/2023\/12\/revolution2.png?resize=1130%2C638&ssl=1\")
<\/a><\/p>\n\n
\n
![\"\"](\"https:\/\/i0.wp.com\/www.smallake.kr\/wp-content\/uploads\/2023\/12\/ai3.png?resize=1095%2C642&ssl=1\")
<\/a><\/p>\n