Hello everyone. We have started building GPT-3 powered decision assistant, would love any feedback.
For the alpha release, the idea is to help with A vs B decisions. The user provides two choices plus context that is relevant for the decision. The assistant provides a recommendation along with the explanation and pros/cons for each choice. (Check out the video on our site)
We want to keep working on it so we can handle the decisions even before the user has narrowed down the choices. We are thinking users could use it for deciding between what to eat (restaurants), products to buy etc.
If there are any use cases that come to mind for this product would love the insight. Cheers.
is this passed openai’s review process?
One of the biggest problems in society is figuring out what to have for dinner. So if you can specialize it to solve that, you’ll probably be a billionaire
I will have my technical co-founder @bradyneal answer the first question in detail, but open ai safety team gave us input filters which we have implemented. We are also implementing an output filter to stop sensitive topics.
The personal assistant because you can provide personal context, I like non-fiction books, I had pizza for lunch with this context, help me decide.
This is a good point we need to do a better job of providing factual information, so we are working on providing sources for the information and working hard to filter false information.
How did you get the OpenAI validation on what seems to be health related advice (intermittent fasting vs. ketogenic: claim that intermittent fasting reduces the risk of heart disease)?
We just tried hard to implement safety stuff before submitting the application and then added what the OpenAI safety specialist asked us to add to our prompt. We actually ended up making it stricter after approval, so we now only accept inputs and outputs that are classified as a “0” by the OpenAI content filter.
I am still trying to understand. It is recommending based on some historical data on your options? It will show you pros and cons and let you decide, Am I right? How would you recommend on something that is new that even GPT is not aware of yet, such as, which vaccine is better for corona?
Yes you are right. It is limited by what the model know, but we are working on using different sources to gather data and show information even if we can make a recommendation.