Google Releases PaLM, a ChatGPT-like ChatBot for Healthcare

• Blockchain Council
• December 30, 2022

MultiMedQA is a revolutionary development in the medical field, offering a powerful language model that combines HealthSearchQA and six existing open-question answering datasets. This disruptive new tool is set to revolutionize the way medical professionals, researchers, and consumers alike access information and answers to their medical queries. With this tool, medical professionals will be able to access information more quickly and accurately than ever before, while consumers will be able to receive answers to their questions in a fraction of the time. This is truly a groundbreaking step towards greater accuracy and accessibility in the medical field.

The model has been designed to evaluate the quality of human responses to complex topics. To accomplish this, it used an elaborate assessment system, which measured the responses’ accuracy, detail, possible detriment, and prejudice. By doing so, it ensures unbiased answers for people and patients to their queries.

The MultiMedQA dataset comprises various datasets of clinical topics, MedQA, MedMCQA, PubMedQA, LiveQA, MedicationQA, and MMLU, as well as a new HealthSearchQA dataset. This HealthSearchQA dataset contains 3375 consumer questions that have been generated from seed medical diagnoses and related symptoms utilizing a search engine. This technology-driven approach allows MultiMedQA to curate frequently asked questions more efficiently.

PaLM, the 540 billion parameter Long-form Language Model, came to the rescue of researchers in their quest to evaluate Long-form Language Models (LLMs) using MultiMedQA. Through its instruction-tuned variation, Flan-PaLM, researchers have achieved state-of-the-art performance on MedQA, MedMCQA, PubMedQA, and MMLU clinical topics. By combining few-shot, chain-of-thought (CoT), and self-consistency prompting techniques, Flan-PaLM has outperformed many strong LLM baselines by a considerable margin. In fact, Flan-PaLM showed over 17% better performance on the MedQA dataset of USMLE questions than the previous state-of-the-art. Despite these impressive results, human evaluation has identified some notable gaps in Flan-PaLM responses.

Med-PaLM was developed with the goal of providing an accurate and efficient method of medical diagnosis. In comparison to Flan-PaLM, Med-PaLM yielded results that were on par with a medical expert’s judgment. A group of doctors assessed that 92.6% of Med-PaLM’s responses were consistent with the scientific agreement, while only 61.9% of Flan-PaLM’s responses were. Furthermore, Med-PaLM responses that could potentially lead to negative consequences were assessed to be similar to those of the clinician-generated answers (5.8%), while those of Flan-PaLM were significantly higher (29.7%). Thus, Med-PaLM became a better and more promising tool for medical diagnosis, as it offered a balance between accuracy and speed.

At the Google for India 2022 event, Google announced a groundbreaking collaboration with Apollo Hospitals in India to revolutionize the use of deep learning models in x-rays and other diagnostic purposes. Through this collaboration, Google hopes to empower doctors and health professionals with cutting-edge technology, allowing them to make more accurate diagnoses and improve patient care. Google has also forged partnerships with many other leading healthcare institutions, including Aravind Eye Care System, Ascension, Mayo Clinic, Rajavithi Hospital, Northwestern Medicine, Sankara Nethralaya, and Stanford Medicine.

As technology continues to revolutionize the healthcare industry, two tech giants are joining forces to bring artificial intelligence to the realm of healthcare. Google and Microsoft are collaborating with OpenAI to utilize GPT-3 to make collaboration between healthcare professionals and teams more efficient. This cutting-edge AI has the potential to make a lasting impact on the industry, modernizing the healthcare experience for clinicians and patients alike.

In November 2022, Meta AI unveiled Galactica, an ambitious AI-generated program with the lofty goal of easing the workload of academic researchers. The program promised to generate comprehensive literature reviews and Wiki entries on any subject, which seemed like a dream come true for the academic community. Unfortunately, the results were unreliable, and the program failed to meet expectations.

At the same time, Meta AI created another groundbreaking AI agent called CICERO that fused natural language processing and strategic reasoning. This AI agent was able to demonstrate an impressive performance in the complex game, Diplomacy. While playing against human players, CICERO was able to exceed the average scores of all other players by more than double ,and was even among the top 10% of players that took part in multiple games. This remarkable achievement of CICERO and other emergent & functional AIs serves as a testament to the power of AI, and its potential to outwit even the most experienced human players.

If you want to keep up with the trends of Blockchain industry, join our communities on Discord, Reddit and Telegram.