AI Progress doesn't feel as fast as we're told (backnotprop.com)

Amid ongoing advancements in AI technology, a developer shares observations indicating a perceived slowdown in significant progress within the AI/ML community. Despite having engaged deeply with AI tools like Claude Code and various agent systems, the individual expresses dissatisfaction with the incremental improvements observed over the past year, suggesting that recent models may not be as robust as previous versions. Concerns are raised over the limitations of long-running agents and the general effectiveness of current automation capabilities, hinting at a potential stagnation in innovation among leading labs. The commentary highlights a crucial pivot towards open-source models, emphasizing their rising importance as proprietary models appear to plateau. The author argues that unless the industry prioritizes transparency and ethical development, it risks producing systems that are incapable of safe and effective AI solutions for the greater good. While acknowledging significant strides in AI for creative tasks, there remains skepticism about the broader implications for humanity, especially regarding alignment and security issues. The call for open-source advocacy signals a critical moment in the evolution of AI technology, as developers seek alternatives to traditional, centralized power structures.