Post #3276819

@kristen_d@mastodon.social @sue@glasgow.social @briankrebs@infosec.exchange The METR study last summer was surprising given the institution's industry alignment; it showed empirically a 19% slowdown by software developers when using LLM-driven assistance. Most of the gains seen probably come from the harnesses around the LLMs; "agentic" is just branding for this. Giving autonomous agency to a nondeterministic non-rational algorithmic entity is all kinds of cybersecurity stupid. They're good at gatling gunning exploits and CVEs though!