Why AI in fintech still needs a human in the loop

Got a little cameo in Forbes India this week

They were covering GPT-5 and its big leap in reducing “hallucinations” by 45–80% (compared to GPT-4o), which is huge for anyone working in AI.

In the piece, I talked about why even with these improvements, you still need a human in the loop when you’re in industries like fintech. Models can ace benchmark tests, but the real test is how they handle messy, high-stakes conversations in the wild.

Curious to see how GPT-5’s numbers hold up when we put it through real-world servicing and collections use-cases.