A recent dinner discussion with business leaders in San Francisco revealed a rift in the AI community over the possibility of achieving human-like intelligence, or AGI, through large language models. While some tech CEOs, such as Anthropic’s Dario Amodei and OpenAI’s Sam Altman, claim that large language models can attain human-level or super-human intelligence, others are more skeptical.
Thomas Wolf, co-founder and chief science officer at Hugging Face, has called the optimistic claims “wishful thinking” and believes that today’s LLMs lack the ability to ask creative questions. He suggests that creating an “Einstein model” – a hypothetical AI system capable of solving complex problems like Albert Einstein – requires deeper understanding of how to get there.
Wolf is not alone in his concerns. Other AI leaders, including Google DeepMind CEO Demis Hassabis and Meta Chief AI Scientist Yann LeCun, have expressed doubts about the potential of LLMs to achieve AGI. They argue that significant technical hurdles must be overcome before such a goal can be achieved.
However, some researchers, like Kenneth Stanley, are working on developing advanced AI models with today’s technology. Stanley believes that creativity is essential for achieving AGI but notes that designing truly intelligent AI requires replicating human subjective taste for new ideas. He argues that the field of open-endedness – which aims to create AI models that can automate scientific innovation – holds promise.
The debate highlights a growing recognition among AI leaders that the achievement of human-like intelligence may be more challenging than previously thought. While some still hold out hope for rapid progress, others are advocating for a more cautious approach and serious, grounded questions about what’s standing in the way of AGI.
Source: https://techcrunch.com/2025/03/19/the-ai-leaders-bringing-the-agi-debate-down-to-earth