Todays chat bots should be considered dumb prototypes, when thinking about even the near future.
Todays chat bots should be considered dumb prototypes, when thinking about even the near future.
They are very clever at transforming data into language, or one language into another... but they work on language and not knowledge.
It would be interesting to see a language model identify the necessary resources to fulfill a query which is then fed back into it to transform the dry data back into language.
Identify if the question is looking for weather, sports scores, unit conversion, general knowlege, or business information for the following requests:
1. "What is the high temperature tommorow?"
2. "Did the packers win last night?"
3. "What time does Walmart open?"
4. "What will it rain next week?"
5. "Where can I buy an umbrella?"
6. "How many feet are in a mile?"
Question classification:
To which GPT responded: 1. Weather
2. Sports Scores
3. Business Information
4. Weather
5. Business Information
6. Unit Conversion
Using this, it should then be sent to a system that knows how to do those queries and return back data.However, doing that integration isn't the place for today's chatbots - or at least not the place for OpenAI to be trying to do all the possible things. Those queries also cost money and become harder to bill for.
Where are you putting the API key to access those services?
Where does it run the code that it wrote and how does it provision those resources?