Gary Marcus in conversation with Will Wilson

AI researcher Gary Marcus is a prominent critic of the scaling hypothesis that drives much of the AI industry today – the belief that simply adding more compute, data, and GPUs will lead to artificial general intelligence (AGI).

Will Wilson, Founder & CEO of Antithesis interviewed Marcus on stage at Bug Bash 2026.

In this interview, Marcus argues that pure large language models are fundamentally limited and will not achieve AGI without incorporating symbolic AI techniques alongside neural networks, an approach called neurosymbolic AI. He points out that major AI companies have quietly begun adopting symbolic methods (for instance, in Claude code) despite publicly maintaining that scaling alone will work. Marcus discusses the economic implications of this shift, suggesting the AI industry may be in a speculative bubble since current LLM-based systems lose money and require constant fundraising based on AGI promises. Marcus lays out the additional breakthroughs he believes will be required for AGI, including better world model construction, temporal and spatial reasoning, and the ability to induce structured representations from raw data – challenges he believes will take at least a decade to solve.

And, of course Marcus provides the kind of predictions he’s become famous for, on capabilities AI will, or will not have by 2027.