There are papers on it from people more knowledgeable than me but what it comes down to in ML is always the dataset. A machine that is already ready to generat

There are papers on it from people more knowledgeable than me but what it comes down to in ML is always the dataset. A machine that is already ready to generate datasets for you improves the quality many times. That means with every Version of ChatGPT, the datasets for other companies grow richer. My personal take is that it takes many AI's in combination with a kubernetes like control loop to reach AGI not a single big one. Take for instance programming. Given enough iteration and therefore money AutoGPT can reach "somewhat usable" code and just like in regular development the whole deal comes down to testing of what you desire. Therefore if you can find a way to "test" your dataset you can iterate automatically and that's easier for some task and more difficult for other but still possible. The common thought is that an AI model needs to "know it all" but as they thought me in business psychology, that's actually not even true for humans themselves. Furthermore, i have mentioned the new "cards" from NVIDIA who are even more powerful than what we had before So yeah i think the space is still wide open.