What is AI ‘model collapse’?

Sam Hawley: Artificial intelligence has come such a long way in a really short time. So, if you actually look up the list, it's quite confronting how many companies have signed up to the OpenAI bandwagon and sort of signed partnership agreements where OpenAI gets access to this company's proprietary data source that they have, and then in return, maybe the company gets access to use some of these tools internally in their business as sort of the cutting edge, if you like. And so, these terms in themselves aren't controversial, but companies like OpenAI, that's right, they are racing to build bigger and better models with the hope of or the goal of producing a generally intelligent artificial intelligence that is able to perform any task a human could perform at the level or better than a human would be able to do that equivalent task. If you then collect that AI generated data and use it to train the next generation of AI models, which is what is happening at the moment, researchers have shown that if that recursive process is too intense, like it's not diluted by human data, or there isn't human data to augment that's AI generated data, then what actually happens over time is the models get worse and worse. Aaron Snoswell: One thing that's clear is that this new technology, generative AI is what people call it, it does seem to genuinely be revolutionising many sort of aspects of how people work and communicate.