Why does ChatGPT make a difference?

In our last meetup on Tuesday, Steffen Brandt looked a bit behind the curtain of current large language models (LLMs) and how they differ from each other, Jan Peter and Matthias provided practical tools and tips on how to use ChatGPT and similar tools efficiently.

The main takeaway from the first presentation is perhaps that while data is still important for training LLMs, it seems that human feedback on generated responses and incorporating this feedback into model fine-tuning will be the more crucial part for future models that want to compete with ChatGPT (see slides here).

Jan Peter Prigge and showed some excellent tools, for example how to integrate ChatGPT into Google Spreadsheets and how to efficiently write individualized marketing mails (the links to the tools are added below).
Matthias Nannt showed the variety of prompts you can use and how to efficiently use the chat history in ChatGPT to let the model benefit from prior knowledge.

Useful links for tools and ideas for good prompts:

PROMTLOOP → https://www.promptloop.com/
CARGO → https://workspace.google.com/marketplace/app/cargo/635545301111
WEBGPT → https://chrome.google.com/webstore/detail/webchatgpt-chatgpt-with-i/lpfemeioodjbpieminkklglpmhlngfcn
MERLIN → https://chrome.google.com/webstore/detail/merlin-chatgpt-plus-app-o/camppjleccjaphfdbohjdohecfnoikec
PROMPTGENIUS → https://chrome.google.com/webstore/detail/chatgpt-prompt-genius/jjdnakkfjnnbbckhifcfchagnpofjffo