4 ms·
I used to be skeptical of the hype but it's hard to deny that they are incredible tools. For coding, they save me a few hours a week and this is just the beginn
by yodsanklai 2y ago
I used to be skeptical of the hype but it's hard to deny that they are incredible tools. For coding, they save me a few hours a week and this is just the beginning. A few months ago I would use them to generate simple pieces of code, but now they can handle refactoring across several files. Even if LLMs don't get smarter, the tooling around them will improve and they'll be even more useful. We'll also learn to use them better.
Also my gf who's not particularly tech savvy relies heavily on ChatGPT for her work. It's very useful for a variety of text (translation, summaries, answering some emails).
Maybe Sabine Hossenfelder tries to use them for things they can't do well and she's not aware that they work for other use cases.
- titanomachy 2y ago> Maybe Sabine Hossenfelder tries to use them for things they can't do well and she's not aware that they work for other use cases. Yeah. That was my first thought. There’s probably orders of magnitude more training data for software engineering than for theoretical physics (her field). Also, how much of software engineering is truly novel? Probably someone else has already come up with a decent solution to your problem, it’s “just” a matter of finding it.