The major research that led to the creation of the AI natural language systems with the capabilities available to the public 2022+ (e.g ChatGPT) has been performed in 2014-2020, it was not a single company work. It was the a many years effort of specialists from different organizations (enterprises, universities and individual specialists), working in different countries.
This are just a few of the landmark events:
- 2014, Generative Adversarial Networks (Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio)
- 2015 OpenAI is created
- 2016, Attention Is All You Need (Dzmitry Bahdanau Jacobs University Bremen, Germany; KyungHyun Cho YoshuaBengio Universit′e de Montr′ eal);
- 2018, BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (Jacob Devlin Ming-Wei Chang Kenton Lee Kristina Toutanova. Google)
- 2020 Language Models are Few-Shot Learners GPT3 related research OpenAI (too many to list)
- 2021 Antrophic (Claude.AI) is created mostly from OpenAI former employees, with an expressed desire of a greater focus on responsible/safe AI
- 2022 OpenAI makes ChatGPT available
Note1: If you don't have patient to check the links, the most relevant point is that the major research that led to the current technology was shared publicity, available to all organizations and countries
Note2: Skipping Lamma3, Qwen2 or Grok from the events for a shorter history.
It is reasonable to assume that since 2020 the published research was sufficient to allow a small group of highly talented specialists in the fields of information technology, data engineering, mathematics and related subjects would be capable of creating an AI model.
- You need a few billion dollars in order to perform: data collection, data curation, compute (training an inference)
- Before the availability of GPT3 in 2020, the expectations around the capabilities of AI models where mostly theorical. Having this together with the investment requirement, it was considered by many as an high risk research.
- The compute for AI requires specialized hardware, NVIDIA invested on the development of this kind of capabilities much earlier then their competitors, however they have a limited production capacity. They cannot satisfy global demand.
- Some countries use their regulatory power to control who has access to the (scarse) NVIDIA hardware
