Large language models - GPT-3 this time

So I’ve started playing with GPT-3 this morning, posing the same questions I’ve asked GPT-J. Seems like the answers are much more coherent. For default configuration parameters I get reasonable stuff, but the model has no concept if something is real or not and makes things up:

– Q: How do I make steel?

• A: There are many ways to make steel. The most common method is to start with pig iron. Pig iron is created in a blast furnace by melting iron ore with coke (a carbon-rich substance made from coal) and limestone (a calcium-rich rock).

– Q: How much carbon is needed to make steel?

• A: One metric ton, or 1,000 kilograms, of steel requires about 1.8 metric tons, or 1,800 kilograms, of carbon to make.

– Q: How much rubber goes into making a car?

• A: It takes approximately 12-15 tires to make a passenger car, and 18-21 tires to make a light truck or SUV.

– Q: Which components of a car are made of glass?

• A: The windshield, window, and mirrors are all made of glass.

– Q: What is pig steel?

* A: Pig steel is a type of cast iron that is made from pigs.

– Q: What is cow iron?

* A: Cow iron is a type of metal that is often used to make tools and other objects. It is a strong metal that is resistant to corrosion and is often used in construction..