May the Source Be with You - The Power of Generative AI in Talent Acquisition

Generative AI has moved past the hype stage, and the recruiters and sourcers who build it into their daily workflow are already pulling ahead of those who ignore it. The argument here isn't that AI replaces recruiters, but that the tools reshape what the job looks like, shifting time away from repetitive tasks like writing Boolean strings and screening resumes toward the harder problem-solving work: why a candidate list lacks diversity, how to fix a broken pipeline, what a hiring manager actually needs.

That shift echoes past disruptions. Calculators didn't kill mathematics; LinkedIn didn't kill recruiting, it created sourcing as a discipline. Large language models are set to do something similar, freeing up capacity for the parts of the job that actually require judgment.

A central idea borrowed from search engine strategy applies directly to LLMs: just as different search engines rank and index results differently, different language models return different answers to the same prompt. Relying on a single model, the way many people default to one chatbot, means missing out on the variety of results that come from trying the same question across tools.

Several free resources make that comparison easy to run:

• Poe, a Quora product that gives access to a wide range of models including GPT, Claude, and open-source options, with the ability to switch models mid-conversation for direct comparison
• Hugging Face, the open-source hub where most publicly available models are hosted, along with Hugging Chat, a free chat interface for testing top models
• Custom assistants built inside Hugging Chat, similar to GPTs, where a model can be given specific instructions and examples to perform a repeatable task, from proofreading to condensing research papers
• LMSYS, a research project out of Berkeley, UCSD, and CMU that ranks models head-to-head through a public chatbot arena

Real-world testing matters because models are not interchangeable. One example involves building a simple time-zone conversion tool: one model handled the calculations cleanly while several well-known alternatives struggled with the same task. Small differences like that add up when a tool becomes part of a daily workflow.

The broader point is that the technology landscape around generative AI keeps expanding, and comfort with more than one model, plus a habit of testing tools against real tasks, is quickly becoming a basic skill for sourcing and recruiting work.