<aside> 💡 This is a living document of notes about the various AI tools out there. I’ll continue to update this as I learn to leverage them to save medical trainees time and energy.
</aside>
Conversational language model chatbot trained with large data set of discoverable online materials (books, forums, articles, websites) up to the year 2021. Master of the language arts and powerful tool for writing, summarizing, or outlining anything you can imagine.
The free ChatGPT research release operates with the language model GPT-3.5, which is OpenAI’s flagship natural language processing chatbot. There are multiple tiers of the chatbot. The most basic is the “Base” model. This is the free ChatGPT version. The most advanced model is “Davinci” and is more powerful in its processing capability and language generation accuracy.
Open AI makes money by allowing companies to integrate their models to power their online services. They get paid by the “token” or length of the prompts and number of prompts asked to the models. They can also be integrated into workflows and trained to perform highly specific tasks. For example, they might want it to answer customer questions in a sassy, semi passive-aggressive manner.
Attention mechanism: method of weighting different parts of an input sequence, allowing the model to selectively focus on the most relevant parts of a prompt while generating a response.
Chat instance: this is the chat window within which the conversation is taking place. You can edit the titles of these chats to keep the projects you work on organized. As of 3/17/24 ChatGPT can now recall information across different chat instances.
Conversation instance: This refers to the actual conversation happening within the chat instance. It includes all the prompts and responses. If you sit down and start talking with the chatbot, you are starting a new conversation instance. A new conversation instance is triggered if the application times-out, system crashes, network outage, error with your program interface, or a prolonged period of inactivity. Also, if you abruptly change topics and start asking it to do something completely different, the context from your previous conversation will be dumped.
Conversational context: This is the frame of reference being utilized by the chatbot to formulate its responses and navigate the conversation. It is stored within its “short term memory” and includes your prompts and its own responses. This is how it determines appropriate responses to new prompts. Every piece of feedback or new prompt you give is interpreted within this context and it is able to refine future outputs. That’s why it feels like you’re talking to a person.
If you start a new conversation instance, the chatbot no longer has the context from your previous instance and you will have to give it the context again and bring it back up to speed. This could include showing it the most recent version of the text you are working on and explaining what you want to do. Because you are essentially starting the conversation over. Ask it what context it is using to answer questions.