This has nothing whatsoever to do with code gpt. Code gpt takes care of the semantics and linguistics but NOT the logic of code, whether strategy code generation or any other program code generation.
The explanation of this is actually quite simple. The trained deep learning function generates output based on what it has trained on and seen. It does not take into account the logic and intent of code. It takes stock of the semantic and linguistic details of code which makes it very suitable for debugging and code issues itself but not the logic and intent of your code. That you need to prescribe outside chatgpt. I currently take care of logic through different flavors of reinforcement learning and the code generation is taken care of by templating and chat gpt. RL is governed by the environment and reward functions that represent to a large degree the intent of your algorithm logic.
Give it a couple of years. The video I posted has a part where it piece together a broken English email into a coherent one based on what it thinks the user intended. Similar thing would apply to code logic. With enough users inquiries and feedback it can improve.
Like seriously, how complex you think retail algo can be? I'm not saying Jim Simmons will start using this for his fund. This is for simple logic codes that 95% of traders without code knowlege would find very useful.
Last edited:
