Is There a Way to Solve the Reliability Problem?

2min Snip

00:00

Play full episode

Summary

Transcript

Episode notes

In his experience with large language models, which has predominantly been through gpt three, there is the tendency of the model to hallucinate. And i think that my question is onow, is there a way to solve that problem? Because that's a major reliability problem. You know, i think the the model needs an estimation of its own confidence, probably in an answer. If that's below some threshold, it should say, i don't know.

Demis Hassabis is one of tech's most brilliant minds. A chess-playing child prodigy turned researcher and founder of headline-making AI company DeepMind, Demis is thinking through some of the most revolutionary — and in some cases controversial — uses of artificial intelligence. From the development of computer program AlphaGo, which beat out world champions in the board game Go, to making leaps in the research of how proteins fold, Demis is at the helm of the next generation of groundbreaking technology. In this episode, he gives a peek into some of the questions that his top-level projects are asking, talks about how gaming, creativity, and intelligence inform his approach to tech, and muses on where AI is headed next.

This is an episode of "The TED Interview," a podcast in the TED Audio Collective. It's hosted by author Steven Johnson. To check out the rest of their episodes, including a recent mini-series on the future of human intelligence, follow the show wherever you're listening to this.