Google never released a state of the art model, what they run in translate, ranking, etc are efficient models on par with state of the art a few years ago. They

Google never released a state of the art model, what they run in translate, ranking, etc are efficient models on par with state of the art a few years ago. They train huge models that are too large to even demo (PaLM 540B) and never release the models, not even for academia to try to build on them. Their voices and OCR are behind competition, I'd rather use ElevenLabs and Amazon Textract. Google makes a lot of noise but delivers little, and what they deliver is almost never the best in the world. I think maybe they made a mistake with TPUs and the PyTorch-NVIDIA path seems to be more productive and can deploy more advanced LLMs. There was a time when I was dying for an invite to Gmail or Google Wave, but that time has long passed and nothing in-between caused that level of excitement. What I appreciate most from Google is YouTube, and not for the UI or recommendations but for simply being the largest video platform with everything you could want in music and education. If I lost access to YouTube internet would be dead for me, well, almost dead.