AI progress may be hitting a wall as hype outpaces real value

A wave of recent AI setbacks, including OpenAI delaying its IPO and scrapping a model over deceptive behavior, signals a broader slowdown in the field. The prevailing belief that ever-larger models automatically bring value is being questioned, as bigger systems tend to generate more convincing falsehoods without a clear path to fixing hallucinations. Meta's new Muse agent, designed for low-information users, has already exposed personal addresses and runs each instance on its own Linux VM, raising concerns about a potential botnet.

Researchers who examined Muse's infrastructure found hard-coded tricks and code resembling other models, suggesting limited originality and safety. While specialized AI applications in science and engineering still show progress, the overall market may be nearing a peak where additional spending yields marginal gains. The article warns that continued hype could mask underlying trust and usability issues, potentially leading to a correction in the sector.