Are you using an LLM to rank your candidates? Chances are you are or will be soon, as every recruitment technology vendor rushes to deploy AI Agents to every level of the hiring funnel. Brainfooder
Martyn Redstone - playing a crucial (and under acknowledged!) role in our community by revealing limitations of the technology. Turns out, a process as venerable as batch processing is endemic in architecture of LLM’s and can produce different candidate recommendations based on
batch size.
Must read folks. Also:
OpenAI theory on why LLM’s hallucinate - our training methods reward binary choices (true / false), so AI ends up preferring to guess, rather than say ‘don’t know’…