Gary Marcus Outlines Seven Predictions for the Upcoming GPT-4 Era
Gary Marcus examines the massive hype surrounding ChatGPT while cautioning that the AI still struggles with reliability and hallucinations. He offers seven predictions for what to expect as the technology evolves into GPT-4.
ChatGPT takes the internet by storm as millions of users flock to experience its uncanny ability to generate human-like text, stories, and poems. People quickly find unconventional uses for the tool, ranging from crafting clever pickup lines for dating apps to writing academic term papers and attempting to reinvent search engines.
Despite the massive hype, Gary Marcus warns that the public remains easily fooled by fluent text that lacks true comprehension. The system frequently hallucinates information, such as absurdly claiming that churros are useful surgical tools, which prompts platforms like Stack Overflow to temporarily ban AI-generated submissions.
Marcus emphasizes that ChatGPT still fails to reliably understand the physical and psychological worlds, no matter how convincing its output appears at first glance. As the AI community looks ahead to the release of GPT-4, he offers seven specific predictions about the future trajectory and limitations of these large language models.