Trend Tuesday: Think RAG Is Yesterday’s News? Meet the Chatbot Milestone Turning Heads
Art by @basilonmypizza: https://lnkd.in/eF8FkWzN - https://basilhefti.ch/
🙂 In 2023, Noy & Zhang (Science, Jul 2023) reported that a GPT-based copilot made knowledge workers «faster, better, happier».
Since then, chatbots have come a long way. They need to be fast. Accurate. Need to handle knowledge base updates. To fence off questions which are outside their scope. Ask for more context. And of course be polite and ethically correct.
Follow-up studies in customer support (Brynjolfsson et al.), consulting (Dell’Acqua et al.), and contract review (Martin et al.) have echoed the same pattern: pairing people with large language models often delivers equal or better quality while saving valuable time ⏱️.
Last week Google reported on AMIE (Articulate Medical Intelligence Explorer) 🩺, a primary-care chatbot. In a randomized, double-blind trial with 159 simulated patients, AMIE matched or outperformed physicians on nearly every clinical and conversational metric 🏅.
The study stands out because it combined broad testing of diagnostic accuracy, communication quality, fairness, and bias 📊 with a self-play training approach: letting the model refine its reasoning autonomously 🧠, a hallmark of DeepMind.
All cases were simulated, however ⚙️. Real-world performance, where patient histories are messy and symptoms overlap, will be lower. This gap marks the next challenge.
Retrieval-Augmented Generation (RAG) may feel familiar and off-the-shelf by now. AMIE illustrates that this journey is only just starting 🚀.
Which domain-specific chatbots should tackle real-world complexity next?
- Noy & Zhang (Science, Jul 2023), https://lnkd.in/e45W_iDc
- Brynjolfsson et al., https://lnkd.in/eh9RrXfY
- Dell'Acqua et al., https://lnkd.in/epRCa9C7
- Martin et al., https://lnkd.in/ekQysTsc
- Tu, T., Schaekermann, M., Palepu, A. et al. Towards conversational diagnostic artificial intelligence. Nature (2025). https://lnkd.in/eJsnQrAg (AMIE)
Follow me on LinkedIn for more content like this.