With GPT-3 you interact with the model, using natural language descriptions, requests, examples and other means to communicate any task. GPT-3 is a huge model, both in terms of data and power. Therefore, it has a qualitatively different behavior.
Unlike other models, if you want to do a new task, it won’t require retraining with additional data. Once the prompt has been adjusted until it understands, the system meta-learns the task based upon the high-level abstractions that it received from pretraining.
This kind of “Prompt Programming” is less like regular programming and more like an coaching exercise. This has similarities to coach athletes: you tell them what they should do and hopefully you’ll get the result you want.
With traditional software that you implement for customers, you have to think things through first. Otherwise it won’t work. With deep learning software, your focus is on providing data that in some way represents the correct answer.
With GPT-3, it’s all about how to describe what you want. In a way, it’s helping to anthropomorphize GPT-3: just like with people, sometimes you get the right answer by just asking the right question in the right format.
One example of GPT-3 capabilities is the emoji transformation. Here, GPT-3 is given a few examples of how to represent film titels as emojis: Back to the Future: 👨👴🚗🕒 Batman:🤵🦇 Transformers:🚗🤖
Want to explore more of what’s possible with GPT-3 for your business? Get in touch with us and follow our tap letter. Add the tap letter to your homescreen.
Mona-Lisa-Steiner-Weg 16/2 1120 Vienna, Austria +43 1 3050836 firstname.lastname@example.org