The report suggests that AI's impact on productivity and economic returns may be overestimated, with significant power demands that could strain utility companies. It highlights a shift in Goldman Sachs' stance, which previously promoted AI's potential for economic growth.

The report includes an interview with MIT economist Daron Acemoglu, who expresses scepticism about AI's transformative impact on productivity and GDP growth. He argues that AI's capabilities are unlikely to significantly improve complex, real-world tasks anytime soon.

The paper criticises the AI industry's marketing narratives, challenging the assumption that more data and computing power will necessarily lead to better AI models. It points out the lack of a clear definition of what "improving AI's capabilities" means in practical terms.

The report discusses the limitations of current AI models, such as GPT, which generate responses based on training data but lack true understanding or intelligence. It raises concerns about the "training data crisis," where improving AI models requires exponentially more data, leading to higher costs.

Acemoglu doubts that large language models (LLMs) will achieve superintelligence and questions the industry's excitement about AI revolutionising various sectors. He warns that premature automation could create more problems than it solves.

Goldman Sachs' Joseph Briggs provides a counterpoint, suggesting that AI could lead to economic benefits by automating tasks and freeing up workers for other jobs. However, the report questions this optimistic view, highlighting AI's challenges in truly automating tasks and creating value.

The report concludes with insights from Jim Covello, Goldman Sachs' Head of Global Equity Research, who is critical of the generative AI bubble. He questions the enormous costs associated with AI and whether it can solve problems that are worth the investment.

Covello challenges the notion that AI costs will decline significantly over time and compares the current AI hype unfavourably to the early days of the internet.

It suggests that the AI narrative may struggle to sustain itself without clear use cases and economic benefits.