Exactly. If someone has already come up with an optimal solution why the hell would I reimplement it. My real problems are not with LLMs themselves but rather the sourcing of the training data and the power usage. If I could use an “ethically sourced” llm locally I’d be mostly happy. Ultimately LLMs are also only good for code specifically. Architecture or things that require a lot of thought like data pipelines I’ve found AI to be pretty garbage at when experimenting