so i guess everyone uses llms to think now - eva's site

Eva, an independent security researcher, published a blog post explaining why she refuses to use LLMs for her vulnerability research, pushing back against the industry narrative that AI is transforming cybersecurity. She said she has never used AI for any research published on her blog, arguing that offloading work to models like Anthropic's Claude would undermine the personal credit and enjoyment central to her work, and that she finds LLM output "confidently wrong" to the point of abandoning sessions. While acknowledging that LLMs are powerful, she contended they still require skilled human guidance — noting that recent headline-grabbing vulnerability discoveries came from experienced researchers at well-funded firms, not from models acting alone, and that an LLM tasked with finding an RCE in the Linux kernel without direction would likely fail. She also criticized AI companies for using cybersecurity as a marketing strategy and for overhyping model capabilities. The post matters because it represents practitioner pushback at a time when OpenAI and Anthropic are marketing frontier models as capable of autonomous security work; watch for whether other independent researchers publicly echo her skepticism and how AI vendors respond to criticism that their security claims outpace real-world performance.