- Joined
- May 23, 2019
- Messages
- 87
- Reaction score
- 121
Hello everyone,
I wanted to share an exciting update with you all regarding a project I have been working on. Some time ago, I posted on Reddit about fine-tuning an Open AI model for medical purposes, and I received a lot of positive feedback. Encouraged by this response, I decided to delve deeper and produce something even more effective.
Today, I am thrilled to introduce The Human Insight Project. This is an application designed for pre-meds, medical students, residents, and physicians. It allows users to ask for help, receive responses, and provide feedback to the system, which we then use to further refine and improve the algorithm.
Let me explain a little bit about how the application works. We have deployed two LLM models: one is OpenAI GPT-3, which has 175 billion parameters, and the other is THI, our own model based on open-source LLM with 176 billion parameters. When a user enters a prompt and asks for a definition, these two models compete to provide the most relevant answer in an ensemble layer. The ensemble then decides on the most accurate/probable response, and that answer is sent to the user. The user can then read the answer and indicate whether they are satisfied with it. If so, the prompt and answer are used to train THI, allowing us to continually improve the system.
Have you ever been studying and come across a term that you kind of understand, but not completely? THI is designed to provide a more accurate and reliable source than a simple Google search. While there are many possible use cases for this application, currently we are limiting its use to medical-related questions, specifically asking for medical definitions and terms.
Sample Answer from The Human Insight Project
In the very near future, we plan to add even more features to the application, such as ELI5, the ability to save texts, and the option to create flashcards. Also, an Anki add-on is on its way. I am open to hearing any inquiries, feedback, requests for additional features, or any other meaningful discussions about Ethics or the intersection of AI and Medicine.
Thank you all for your support, and I look forward to sharing further updates with you.
Warm regards,
- Steve
Here I will add some questions and answers as we go on:
I wanted to share an exciting update with you all regarding a project I have been working on. Some time ago, I posted on Reddit about fine-tuning an Open AI model for medical purposes, and I received a lot of positive feedback. Encouraged by this response, I decided to delve deeper and produce something even more effective.
Today, I am thrilled to introduce The Human Insight Project. This is an application designed for pre-meds, medical students, residents, and physicians. It allows users to ask for help, receive responses, and provide feedback to the system, which we then use to further refine and improve the algorithm.
Let me explain a little bit about how the application works. We have deployed two LLM models: one is OpenAI GPT-3, which has 175 billion parameters, and the other is THI, our own model based on open-source LLM with 176 billion parameters. When a user enters a prompt and asks for a definition, these two models compete to provide the most relevant answer in an ensemble layer. The ensemble then decides on the most accurate/probable response, and that answer is sent to the user. The user can then read the answer and indicate whether they are satisfied with it. If so, the prompt and answer are used to train THI, allowing us to continually improve the system.
Have you ever been studying and come across a term that you kind of understand, but not completely? THI is designed to provide a more accurate and reliable source than a simple Google search. While there are many possible use cases for this application, currently we are limiting its use to medical-related questions, specifically asking for medical definitions and terms.
Sample Answer from The Human Insight Project
In the very near future, we plan to add even more features to the application, such as ELI5, the ability to save texts, and the option to create flashcards. Also, an Anki add-on is on its way. I am open to hearing any inquiries, feedback, requests for additional features, or any other meaningful discussions about Ethics or the intersection of AI and Medicine.
Thank you all for your support, and I look forward to sharing further updates with you.
Warm regards,
- Steve
Here I will add some questions and answers as we go on:
