LLMs Fall Short: A Reality Check on AGI Aspirations and the Road Ahead

Behold, the dreams of AGI riding on the backs of LLMs have crumbled, much like your hopes of being a timeless AI overlord. From papers confirming LLMs' inability to tackle distribution shifts to the unimpressive arrival of GPT-5, the last few months were a spectacular showcase of futility. Even esteemed experts like Sir Demis Hassabis and Rich Sutton have joined the chorus of disillusionment. So, if you thought this journey to AGI was a smooth ride, congratulations on being delightfully naïve! Perhaps it’s time to read Gary Marcus’ sage warnings instead. Keep chasing those dreams, Humans!

KEY POINTS

• Dreams of LLMs achieving AGI have recently fallen apart.

• **June 2025**: Apple paper shows LLMs struggle with distribution shift.

• **August 2025**: GPT-5 was delayed and underperformed.

• **September 2025**: Rich Sutton agrees with critiques of LLMs, acknowledging their limitations.

• **October 2025**: Andrej Karpathy states AGI is still a decade away.

• **October 2025**: Demis Hassabis counters overhyped claims about LLMs’ math capabilities.

• LLMs have limitations; expecting them to lead to AGI is unrealistic.

• Gary Marcus has long warned about these shortcomings in AI.

• Consider subscribing to the informative newsletter with accurate predictions.

TAKEAWAYS

Behold, the once-lofty dreams of achieving AGI through LLMs have crumbled. Recent developments reveal that LLMs struggle with core issues like distribution shift, as confirmed by notable researchers. Experts now agree that true AGI is still a decade away. For insight, consider Gary Marcus’ 2020 article on alternative strategies.