Person
1 episodes · 13 quoted claims
Aishwarya Reganti cites an Air Canada incident in which an AI agent hallucinated a refund policy that was not in the airline's playbook.
“Air Canada had this thing where, um, one of their agents predicted or hallucinated a policy, um, for a refund, which was not part of their original playbook.”
Reganti says a paper from UC Berkeley and Databricks researchers found about 74 to 75 percent of enterprises surveyed named reliability as their biggest problem, which made them hesitant to deploy AI products to users.
“about seventy four, seventy five of the enterprises that they had spoken to. Um, their biggest problem was reliability, and that's also why they weren't, uh, comfortable, um, deploying products to their end users”
Reganti says the CEO of Rackspace blocked time from 4 to 6 am each day to catch up on AI.
“I used to work with the CEO of Now Rackspace, Gajen, so he would, um, have this block every day in the morning, we should say catching up with AI, four to six am.”
Reganti says GPT-4o is being deprecated in OpenAI's APIs, so companies must move to GPT-5, whose different behavior can throw off a system's calibration.
“GPT 40 doesn't exist anymore, or it's going to be deprecated in APIs as well. So most companies that were using 40 should switch to 5, and 5 has very different properties.”
Reganti says that after about three months, underwriters at a bank using an AI loan-review assistant began asking deeper, unanticipated questions, requiring the system to be rebuilt in part.
“post three months, we realized that they were so excited with the product that they started asking very deep questions that we never anticipated.”
Reganti says a fully autonomous customer-support agent she worked on was shut down because the team could not keep up with the hot fixes for emerging problems.
“we had to shut down the product because we were doing so many hot fixes and there was no way we could, um, count all the emerging or, uh, emerging problems”
Reganti estimates that replacing a critical workflow or reaching significant ROI with AI takes four to six months of work, even with a strong data and infrastructure foundation.
“To replace any critical workflow or to, um, build something that can give you significant ROI, easily takes four to six months of work, even if you have the best data layer and infrastructure layer.”
Reganti says Martin Fowler coined the term 'semantic diffusion' in the 2000s for a term whose meaning blurs as people apply their own definitions.
“Martin Fowler at some point had this term called semantic diffusion back in the 2000s.”
Reganti says retail companies' product category hierarchies are often messy, which makes even the simple task of routing support tickets hard for an agent.
“Any retail company, any popular retail company that you can think of, has hierarchical taxonomies. Most of the times, the taxonomies are incredibly messy.”
Reganti describes insurance pre-authorization of items like blood tests and MRIs as a good AI use case, while invasive surgery should keep a human in the loop.
“Insurance pre authorization is a very ripe use case for AI because, uh, clinicians spend a lot of time, um, pre authorizing, uh, things like blood tests, MRIs and things like that, right?”
In the continuous development half of the guests' framework, a team scopes the capability and curates a dataset of expected inputs and outputs before building.
“we have this right side of the loop, which is continuous development. Uh, where you scope capability and curate data, essentially get a data set of what your expected inputs are”
Reganti says she and Badam maintain a free GitHub repository of AI learning resources with about 20,000 stars.
“We also have a GitHub repository with about 20K stars, and that repository is all about good resources for learning AI.”
Reganti recommends 'When Breath Becomes Air', a memoir by neurosurgeon Paul Kalanithi, who was diagnosed with lung cancer at about 31 or 32.
“it was written by Paul Kalanithi. I think he was, um, um, an Indian origin neurosurgeon who was diagnosed with lung cancer at thirty one or thirty two”