OpenAI recruits hundreds of providers in different regions of the world to improve the code of ChatGPT, mentions Semafor.
Specifically, the company is hiring computer programmers to create training material that will include lines of code, as well as explanations of the code written in natural language.
OpenAI already has a ChatGPT model specifically for translating natural language into code, called Codex. Prior to its launch in 2021, Codex was trained on data pulled from GitHub, the Microsoft-owned code repository.
Building “a system capable of anticipating the effect of its own actions”
It’s good enough that Microsoft uses Codex to power GitHub Copilot, a service that helps programmers write code.
Recent research, however, suggests that programming support could be improved by adding ChatGPT’s unique ability to interact with humans. So it’s no surprise that OpenAI is investing in this area by hiring more developers. Meanwhile, Stack Overflow, the Q&A site for programmers, has banned ChatGPT-generated answers because even its low-quality answers can seem plausible.
According to Yann LeCun, chief scientist of AI at Meta, to really advance a programming assistant, researchers would have to build “a system capable of anticipating the effect of its own actions, and have “a kind of internal model of the world, a mental model of how the world will change as a result of his own actions”.
Copilot is really like a copilot!
LeCun recently compared development assistants like Copilot to cruise control in cars. “Your hands should stay on the wheel at all times”, because Copilot can generate errors in the code without his being aware of it.
“The question is how do you deal with systems that generate code that sometimes works and sometimes doesn’t,” LeCun said. “And the answer to that question is that all of these systems today are not capable of planning; they are completely reactive.”
Source: “ZDNet.com”