Many-Shot In-Context Learning

Rishabh Agarwal, Avi Singh, Lei M. Zhang, Bernd Bohnet, Luis Rosias, Stephanie Chan, Biao Zhang, Ankesh Anand, Zaheer Abbas, Azade Nova, John D. Co-Reyes, Eric Chu, Feryal Behbahani, Aleksandra Faust, Hugo Larochelle – 2024

Basically, use all the examples you can within the prompt. Performance doesn’t max out until you get to pretty large scale, on the order of hundreds or thousands of example. In other words, if accuracy is your main concern, you should use as much of the context window as you can.


References

paperreadonline