3 ms·
I consider code to be non-fiction, and ChatGPT will generate stuff that compiles and works most of the time. There's no need to triple-check the code output.
by copperx 3y ago
I consider code to be non-fiction, and ChatGPT will generate stuff that compiles and works most of the time. There's no need to triple-check the code output.
- brk 3y agoBut this is not really an accurate comparison. "Code" is really a much much much smaller and much much much more structured output than "English words". Presumably, the system was trained with a very small amount of "untrue code" in the sense of stuff that just absolutely could never work. And also presumably, it was trained with a lot of free-form text that was definitely wrong or false, and highly likely to have been originally created to be purposefully misleading, or at a minimum, fiction. That the system outputs reliable code tells us nothing about its current ability to output highly reliable free form text.
- gAI 3y agoBut it understands English words and translates the meaning behind them into code very well, particularly with a bit of iteration. Simply taking the interface from writing code to speaking in plain language is a huge practical accomplishment.
- gAI 3y agoWith the Code Interpreter beta active, it will compile and run that code in the same chat window.