As a dev, there's a basic rule I've heard a lot, that, "don't re-invent the wheel".<p>If you lookup at ChatGPT or any other large research on AI, considering the significant monetary investment, number of hours of work by employees and staff, running the models on high end servers, costing higher emissions, the whole chain of events. Just so we can talk to computers as we'd usually talk with fellows? We're still the one initiating the conversation. We're and will be still the one for whom response is provided for. The only species that matters here is currently only humans, can't we just compromise? Enter and read inputs in a form computers understand and respond?<p>So, can't we just make do with previous forms of communication with the computers? What wall are we really facing that doesn't require compromise from us in one way or other, so that for it the solution is an LLM?<p>In terms of the approach taken for building LLMs is not actually much innovative than brute force, last extraordinary work I'd say was Transforms and OCR, hence a model like BERT would have been evolved into and LLM that we're all talking about today, but considering all the efforts and the tradeoffs we face is it still worth it in your opinion?<p>Is it that much of an improvement?