CB On Air

There are genuine operational details scattered throughout - the three-phase LLM roadmap, the Georgian-language trade-off, and the suptech model validation framework - but the episode is padded with generalities about AI literacy, cultural transformation, and ends on a platitude. Useful signals but not densely packed.

“either uh, data confidentiality, data safety if you use open source models or uh, data quality generated quality. So if you use open source model they generally do not work in Georgian because Georgian is a low resource, uh language”

“we interacted with all 26 departments within the central bank and um asked uh the most relevant use cases for them”

The Georgian low-resource language creating a concrete dilemma between open-source safety and proprietary quality is a genuinely non-obvious and specific insight. Everything else - AI sandboxes, model risk frameworks, bottom-up then top-down adoption - follows the standard central bank AI playbook with little contrarian framing.

“we face a trade off. So either uh, data confidentiality, data safety if you use open source models or uh, data quality generated quality”

“For me, it's a marathon, not a sprint. It's a journey. And, uh, if you consider marathon like 26 miles, I would say somewhere in the fifth mile”

Both guests are genuine hands-on practitioners - an AI engineer building and deploying the actual tools and the department head setting regulatory strategy - not career podcast guests or abstract thought leaders. The National Bank of Georgia is a smaller institution but the guests speak from direct operational experience.

“The platform was in the testing phase for several months and now it has went live, uh, it's been live for several months now and it's being actively used by our employees on a daily basis”

“we do use AI assistance for coding but you have to be careful in that process. So AI generated code contains vulnerabilities, uh, often a lot of errors and it might cause data leakage”

A handful of concrete figures appear - 80% market share concentrated in two banks, 26 departments consulted, 10 tools live and 10 in development - but there are no outcome metrics, cost figures, processing-time reductions, or case-level results from any deployed tool. Claims about efficiency and cultural transformation go entirely uncorroborated.

“80% of market share belongs to two major quite agile Georgian banks”

“we interacted with all 26 departments within the central bank”

The host shows some genuine curiosity with relevant follow-ups on Anthropic geopolitical risk, token costs, and AI-generated code vulnerabilities, which are better-than-average questions for this format. However, no claims are challenged, answers are accepted without probing for evidence or concrete outcomes, and the closing question invites pure platitude.

“with everything that we've seen with Anthropic for example recently where foreign users have been locked out how are you addressing those kind of geopolitical risks”

“I also wanted to ask about the costs. How are you factoring in the costs of using tokens which seem to be growing exponentially”