How difficult will it be to train and build an AI Agent that has expertise in a given domain? Will it happen in the next year, or 3 years or 10 years? And who will benefit in the marketplace from his evolution?

SHOW: 984

SHOW TRANSCRIPT: The Cloudcast #984 Transcript

SHOW VIDEO: https://youtube.com/@TheCloudcastNET

CLOUD NEWS OF THE WEEK: http://bit.ly/cloudcast-cnotw

CHECK OUT OUR NEW PODCAST: "CLOUDCAST BASICS"

SHOW SPONSORS:

• [Mailtrap] Try Mailtrap for free
• [Interconnected] Interconnected is a new series from Equinix diving into the infrastructure that keeps our digital world running. With expert guests and real-world insights, we explore the systems driving AI, automation, quantum, and more. Just search “Interconnected by Equinix”.

SHOW NOTES

• OpenAI looks to train their models to replace junior bankers

WHAT WOULD BE THE STAGES OF AN EXPERT AGENT?

• Train it on a set of standard knowledge (e.g. Masters of Accounting, Auditing, International Tax)
• Train it on a set of well-defined case studies, to provide industry context
• Train it on a set of adjacent case studies and other domains (business, law, specific industries)
• How to train corner cases?
• How to train gray areas like ethics, morality, or cost-benefit analysis?
• Who is motivated to train these experts?
• What would the cost of these experts be? Can it be similar to a human, or need to be a fraction, or a premium?
• Is there a way to build memory (e.g. experience) without disclosing client information?
• Is there a way to build shareable knowledge between agents for reinforcement training/learning?

FEEDBACK?

• Email: show at the cloudcast dot net
• Twitter/X: @cloudcastpod
• BlueSky: @cloudcastpod.bsky.social
• Instagram: @cloudcastpod
• TikTok: @cloudcastpod