philip lelyveld The world of entertainment technology

The Unexpected Winners Of The ChatGPT Generative AI Revolution

... Getting GAI systems to produce exactly what you are looking for with all of the generation conditions met is not easy. There are many ways to improve the accuracy algorithmically without having access to the prior experimental data. And the power of the algorithmic approach including simulation and zero-shot learning should not be underestimated. Those, who think that AI is all about training data, and, believe me, most pharma company executives do, should refer to the Infinite Monkey Theorem. If you take the infinite number of monkeys with the infinite number of typewriters, eventually, you will get every book that is ever written. And, by extension, if you can efficiently train these monkeys to produce accuate and useful output, you can potentially generate even better books. ...

But someone needs to validate the output. And while it is possible to use both humans and algorithms to evaluate the quality of the text, voice, and images, when it comes to biology, chemical, and experimental physics data, only algorithms trained on this specific data and, in some cases, expert humans, can be good arbiters for generative AI. This makes very specialized high-quality data and expert humans very valuable for training of generative AI systems. ...

The value of high-quality human-generated content, especially the content rich in verified facts, just increased exponentially.  ...

The developers of generative AI will require high-quality data and expert humans for training of the generative and predictive models as well as for benchmarking and validation. And since much of this content is paywalled, it is very likely that the following publishers will be the greatest winners in the generative AI revolution.  ...

See the full story here: https://www.forbes.com/sites/alexzhavoronkov/2023/02/23/the-unexpected-winners-of-the-chatgpt-generative-ai-revolution/?sh=3ddd7c6d12b0