Is AI progress exponential? Anthropic's own research disagrees (www.resultsense.com)

Anthropic finds itself in a contradiction this month as its CEO, Dario Amodei, claims that AI is progressing exponentially towards “Powerful AI,” while the company’s own research indicates otherwise. Amodei's essay emphasizes that AI's scaling laws suggest rapid increases in cognitive capabilities with more computing power, predicting significant advancements within one to two years. In contrast, internal system cards for Anthropic’s latest models, Claude Mythos and Fable 5, report that any identified improvements stem from human research rather than self-acceleration of AI technology. Notably, they found no evidence of the projected "2× acceleration" that would characterize exponential growth. This discrepancy highlights crucial considerations for the AI/ML community and businesses relying on AI advancements. While models are improving steadily, Anthropic’s findings suggest that there may be limitations to rapid progress, pointing to a potential flattening growth curve. For decision-makers, the lesson is clear: invest cautiously, aligning strategies with verifiable improvements rather than ambitious forecasts. By focusing on current capabilities and treating future projections as optional rather than predetermined, businesses can navigate the evolving landscape of AI more effectively, minimizing risks associated with overestimating the technology's trajectory.