Skip to content
TopicTracker
From garymarcus.substack.comView original ↗
TranslationTranslation

The month Generative AI lost its mojo

The article discusses how June 2024 marked a turning point for generative AI, with notable developments and challenges emerging that undermined the technology's earlier momentum and promise.

Background

Gary Marcus is a prominent AI researcher and critic known for his skepticism about deep learning and large language models. This piece was published in late June 2024, during a period when several high-profile generative AI products and companies faced public setbacks, fraud allegations, and financial challenges. Key events included: the collapse of Stability AI (whose image generator Stable Diffusion sparked the generative AI boom), legal battles over training data copyright, the implosion of AI-powered search startup Perplexity amid plagiarism accusations, concerns that OpenAI's GPT models were hitting performance plateaus, and growing investor skepticism about the massive capital expenditures required for generative AI. Marcus has long argued that generative AI systems are fundamentally unreliable and that the hype cycle would inevitably deflate once investors and users encountered their limitations — this post is his declaration that the inflection point has arrived.

Related stories

  • Humanity isn't ready for the coming intelligence explosion

    8.0

    Humanity is unprepared for the rapid acceleration of artificial intelligence that will soon surpass human capabilities. Current safety measures and governance are insufficient to manage an "intelligence explosion," risking catastrophic outcomes. Urgent global coordination is needed for safe AI development.

  • Hardware Is Asynchronous. Most of Our Operating Systems Still Aren't

    4.0

    The article argues that while modern hardware operates asynchronously with multiple independent components, most operating systems remain fundamentally synchronous in design. This mismatch leads to inefficiencies and complexity, as OSes must layer async abstractions on top of inherently async hardware. The author calls for rethinking OS architecture to embrace asynchrony natively.

  • The Humanoid That Runs Until Someone Steps In

    3.0

    The article discusses the development of humanoid robots designed for continuous operation until human intervention is required, highlighting advancements in autonomy and endurance for real-world applications.

  • Ethereum: Still Chasing Sovereignty

    3.0

    Ethereum continues its pursuit of sovereignty as a decentralized blockchain platform, reflecting ongoing efforts to achieve greater independence from centralized control and external influence in its network operations and governance.

  • Developers don't understand CORS (2019)

    3.0

    Many developers misunderstand CORS, believing it is a server-side security measure. In reality, CORS is a browser-enforced policy that restricts frontend JavaScript from reading cross-origin responses, but does not block requests from non-browser clients or server-to-server communication.