Re:Low-code, no-model recommendation system with GPT3
by PeterStuer on Wednesday September 09, 2020 @02:59AM
(#24418089)
Nice as a stunt, but feels like taking a 40 ton truck to get your lunch sandwich from the bakers.
[ Parent
| on HN ]
Re:Low-code, no-model recommendation system with GPT3
by feral on Wednesday September 09, 2020 @05:26AM
(#24419076)
If this turns out to be easy and convenient, it won't matter whether the tech is overpowered.
A family car is really overpowered for most trips to the bakers.
Yet people take the car all the time, instead of walking or cycling, because it's just more convenient and easy, and people are lazy.
You can't write just this off as a stunt.
If you described how YouTube renders cat videos to Tim Berners Lee in the mid 90s, it would also seem like a 40-ton truck.
And yet here we are.
Re:Low-code, no-model recommendation system with GPT3
by lukasLansky on Wednesday September 09, 2020 @05:25AM
(#24419073)
I guess it depends how you think about it. From computational perspective, it's heavyweight. On the other hand, it really optimizes for minimal human effort, on several levels:
* GPT-3 is conceptually simple: the effort to invent it and engineer it amortizes well.
* Once the model is trained, it can be reused to very different tasks.
* Querying doesn't require programming knowledge.