An AI counseling bot that keeps learning after it's deployed, instead of freezing once training ends.
Researchers behind a new system called PsyEvo built a chatbot therapist with three moving parts. One tracks what each client responds to and updates a per-client profile after every session. A second refines how the bot phrases its interventions, pulling in feedback across many different clients' conversations. A third checks that feedback is consistent before it gets used to adjust anything. In tests against simulated clients, the full system scored 7.684 on a benchmark called PsychEval, and stripping out any of the three parts dropped the score by roughly 0.14 to 0.17 points.
That's a real distinction from most AI therapy tools, which ship as fixed models that never adjust after launch. The world has far more people needing mental health support than it has licensed therapists, and that gap is the whole premise this kind of tool is chasing. Letting a bot personalize its approach per client, and get collectively smarter across clients, is a plausible way to make an LLM counselor feel less generic over time.
Still, simulated clients and a bespoke benchmark aren't the same as real patients and clinical outcomes, and a 7.684 score means little without knowing the scale it sits on. Promising architecture, unproven therapy.