so i guess everyone uses llms to think now (eva.ac)

A recent blog post from a security researcher highlights a personal stance against using large language models (LLMs) in research, emphasizing a preference for hands-on coding and problem-solving. The author expresses discomfort in attributing credit to AI, viewing their work as a personal endeavor rather than a collaborative effort with machines. Their stance underscores a belief that while LLMs are powerful, they require substantial human guidance to effectively tackle complex tasks, such as discovering vulnerabilities in software. The post critiques how some in the AI and cybersecurity fields promote models as panaceas, often overstating their autonomous capabilities. Significantly, the discussion reveals a growing debate in the AI/ML community on balancing the benefits of automation with the intrinsic joys of personal exploration and craftsmanship in software development. The author points out that recent vulnerabilities have mostly been found by skilled humans, not solely through AI. Furthermore, the piece critiques the marketing strategies of AI companies that glorify automated vulnerability discovery while neglecting responsible disclosure practices. This perspective encourages professionals to continue engaging in the learning process that accompanies hands-on work, suggesting that the journey of discovery is just as valuable as the end result.