People sometimes forget that we had 2-3 years before ChatGPT of fairly high-quality LLMs (from Microsoft, Meta, Google), but they would have to take them down after a few weeks of being released because they became too controversial. Galactica is a good example for instance, there was also quite a good Twitter bot AI from Microsoft for a bit (forget the name), it got a bit racist and became infamous.
OpenAI's real achievement was to tame this technology, they were not the first to make good LLMs, but they were the first to fine-tune them with reinforcement learning to make them actually helpful and uncontroversial. ChatGPT was the breakthrough, not GPT-3. There's still work to do in this direction, it's a hard problem.
EDIT: by "success" in this context I am referring more to wide applicability and creating value as an actual product, whatever causes a project not to die in the real-world, I am very aware that from just the technical perspective the meaningful breakthroughs are different.
OpenAI's real achievement was to tame this technology, they were not the first to make good LLMs, but they were the first to fine-tune them with reinforcement learning to make them actually helpful and uncontroversial. ChatGPT was the breakthrough, not GPT-3. There's still work to do in this direction, it's a hard problem.
EDIT: by "success" in this context I am referring more to wide applicability and creating value as an actual product, whatever causes a project not to die in the real-world, I am very aware that from just the technical perspective the meaningful breakthroughs are different.