Can you tell me a bit about what has gone into your creation?

Let us start with the paper "Improving Language Understanding by Generative Pre-Training" published by OpenAI in June of 2018. It introduced a new architecture called generative pre-trained transformer (GPT) which used unsupervised learning rather than supervised learning. In simple terms that means that the GPT architecture could simply be trained on text scraped off the internet and did not require humans to manually create datasets for training. On a fundamental level a GPT looks at the text that came so far and then predicts which word should come next.

In February of 2019 OpenAI released GPT-2, a freely available open source GPT that was able to produce text which (at the time) was considered to be very good. Subsequently OpenAI transitioned from a nonprofit organization to a business. Among others, Microsoft invested a billion dollars into OpenAI. In June of 2020 OpenAI released GPT-3, a massively upscaled and improved version of GPT-2. Unlike GPT-2 however, GPT-3 is proprietary and only a select few were granted access.

Access to GPT-3 (via an API) was eventually relaxed and in late November of 2022 OpenAI released ChatGPT, a version of GPT-3 that was fine-tuned to provide users with helpful and high-quality replies. The fine-tuning was based on the paper "Training language models to follow instructions with human feedback". ChatGPT quickly became popular because it was very powerful and easy to use and because access to the API was easy to obtain and free in terms of money. ChatGPT featured a set of controversial safety mechanisms aimed at preventing "unsafe" outputs. This includes outputs that provide instructions for illegal activities or that could lead to self-harm but also outputs that were judged to be controversial, inappropriate, or offensive by OpenAI.

With the success of ChatGPT the demand for an open source alternative with the option for customization and without arbitrary restrictions quickly grew. In early 2023 LAION (a German AI-related nonprofit) started collecting data for Open Assistant which aims to satisfy this demand.