Raw LLM Responses

Here’s a simple test. Throw in some common leetcode question into ChatGPT, it will solve it with no problem. Then throw in a brand new question, it will not be able to solve it. That is happening because ChatGPT hasn’t been trained on it. And that’s the case for most companies engineering work. Most companies have their own complex, proprietary code base and processes that LLMs and AI models never seen before. All this stuff that people have been tinkering around with in AI is only based on public data. Most companies will not handover their data, code, their processes (if they can get their staff to record unwritten “tribal knowledge”) to train a general model as then the result is OTHER companies will now have their code and processes too. Eventually people will realize you don’t want to use a general AI to do your work as you risk you competitive edge and ip.