My issue is not that is just autocomplete (although I truly believe we can do better and is the wrong direction) my issue is the training aspect of the technology, you train massive models on bilions even trilions of tokens and once that model is trained it is done, it is released and is the next big thing yes it can get tools to search the web but it cannot self update it cant modify its own weight it is pre trained on that model is done if more information or new data comes in to get it into its actual neuronetwork we need to now come out with another model model 4.9 or 5.7 or where ever they are in each company. Now I know it is mitigated with RAG and external or adjacent databases for memory but this seems like a bandaid anything happens to that database is gone and even if it didnt the core selling point and core aspect of the tech which is neural network is static. If we can figure out a way to get a model to self update weight in a easy way because now it is nearly impossible with current tech is the very issue with fine tuning, you cannot add without removing or replacing. Then even if is autocomplete I would almost or entirely say fine AGI is here.
What if it’s not just autocomplete?