Why You Should Wait Out AI’s Super-Spending False Start

AI investment surge may face diminishing returns as large language models hit fundamental limits, warns Janusz Marecki, CEO of Fractal Brain and AI partner at Ahren Innovation Capital. Current LLMs struggle with a "data ceiling," where quality training datasets are exhausted, curbing performance gains despite increased compute power. Scaling compute yields shrinking improvements, with costs rising exponentially while accuracy plateaus, raising questions about long-term viability. Persistent issues like hallucinations and probabilistic errors remain unresolved, undermining reliability for critical applications despite massive investment. Marecki advises caution, suggesting the AI boom’s early phase may be overhyped, urging patience for breakthroughs beyond current LLM architectures.