The two primary challenges preventing GPT from replacing programmers are:<p>1) AI lacks the creativity and critical thinking ability of human programmers.<p>2) The 8,000 token limitation of code-davinci-002.<p>Regarding these points:<p>While top 1% software engineers may handle more complex tasks, many programmers mostly perform repetitive tasks such as finding npm modules, importing them, reading documentation, testing, and debugging. Product vision is typically led by product owners or managers.<p>The token limitation is a significant issue. Storing an entire codebase in a language model is not feasible, as it would exceed the token limit. My team at MarsX has addressed this problem by creating a higher level abstraction programming language, which requires fewer tokens, allowing an entire app to be stored within a context. We plan to release this on GitHub soon.<p>Given these challenges, I believe that many programmers will likely lose their jobs within 1-3 years.<p>Is there anything else I should consider?
> Given these challenges, I believe that many programmers will likely lose their jobs within 1-3 years<p>It depends. If you're a programmer, you would be naturally skeptical of an AI dictating how a program should behave. If a human isn't in the loop: foot, meet gun.