Hassabis flags irony in ChatGPT’s ad ambitions

Google DeepMind chief executive Demis Hassabis has challenged the way artificial general intelligence is being discussed and marketed, pointing to what he described as a “big irony” in OpenAI’s plans to introduce advertising into ChatGPT while simultaneously framing the system as a step towards human-level intelligence. His remarks underline widening philosophical and commercial differences between leading AI labs as competition intensifies and public scrutiny grows.

Speaking in a series of public forums and interviews over the past year, Hassabis has questioned claims that systems such as ChatGPT are close to achieving artificial general intelligence, or AGI, a term widely used to describe machines that can match or exceed human cognitive abilities across a broad range of tasks. He has argued that true AGI would need to demonstrate original scientific insight, creativity, and the capacity to generate genuinely new knowledge, rather than rely on pattern recognition and probabilistic text generation.

The comments gained renewed attention after OpenAI signalled plans to explore advertising as part of its long-term revenue strategy for ChatGPT. For Hassabis, this highlighted a contradiction between portraying a system as a future form of general intelligence and monetising it in ways similar to conventional digital platforms. He suggested that if AGI were truly imminent, its value and implications would extend far beyond ad placement or subscription models.

Irony of ads versus AGI ambitions was how the Google AI chief framed the tension, noting that the pursuit of short-term commercial returns risks diluting the seriousness of the AGI concept. In his view, labelling current large language models as AGI, or even as being on the cusp of it, risks misleading policymakers and the public about both the capabilities and the limitations of the technology.

Hassabis has consistently maintained that AGI remains at least five to ten years away, even under optimistic assumptions. He has said that while today’s models show impressive fluency and breadth, they still lack a deep understanding of the world, persistent memory, and the ability to reason robustly outside the data they were trained on. Breakthroughs in areas such as causal reasoning, planning, and learning from minimal data are, in his assessment, still required before AGI can be credibly claimed.

OpenAI chief executive Sam Altman has taken a more expansive view, often describing AGI as a continuum rather than a single moment of arrival. Altman has suggested that systems can be considered increasingly “general” as they improve, a framing that allows for incremental progress while still invoking the broader goal. Critics, including Hassabis, counter that such flexibility risks turning AGI into a marketing label rather than a rigorous scientific benchmark.

The debate reflects deeper differences in how leading AI organisations balance research ambition with commercial pressure. Google DeepMind, which operates within a large advertising-driven parent company, has emphasised long-term research milestones and scientific credibility, particularly after its successes in protein structure prediction and game-playing systems. OpenAI, by contrast, has leaned heavily into rapid deployment, partnerships, and consumer adoption, with ChatGPT becoming one of the fastest-growing software products in history.

Industry analysts note that advertising discussions around ChatGPT come amid rising costs associated with training and running large models. Compute expenses, data centre investments, and talent competition have pushed AI labs to seek diversified revenue streams. Subscriptions alone may not be sufficient to sustain the scale of ambition outlined by companies pursuing ever larger and more capable systems.

At the same time, regulators and policymakers are paying closer attention to how AI capabilities are described. Overstating progress towards AGI could influence regulatory debates, public expectations, and investment flows. Hassabis has warned that hype cycles risk backlash if systems fail to live up to inflated claims, potentially undermining trust in the field as a whole.