2 ms·
I just told it to generate an HTML page listing 10 science fiction novels, linking to Amazon and listing the ISBN and price for each. It just worked. I told it
by krapp 4y ago
I just told it to generate an HTML page listing 10 science fiction novels, linking to Amazon and listing the ISBN and price for each. It just worked. I told it to write a small C program to open a window with SDL2. It worked, although it didn't include the event loop, which it might have if I specified. People are already using it to generate code, check logs, and automate all sorts of other tedious tasks, you can just check HN to find lots of examples of people finding legitimate use cases for LLMs.
I mean, I get a lot of the moral arguments against it - particularly where image generation is concerned - but we're well past the point of wondering whether it's even actually useful, it just isn't useful for everything at the moment. But yes, it is useful. It doesn't have to be perfect - humans aren't perfect. Humans "hallucinate" working code and truthful data all the time. But something that's adequate for a lot of tasks and which will get you most of the way there for a lot of other tasks is still useful.
Edit: I asked it to generate an svg image of a smiling cat face against a white background and... it wasn't so great. I asked for one of the Nike logo and it seems to have hallucinated an external link to a wordpress site. I told it to only use valid urls and it gave me this:
<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="200" height="200">
<rect width="200" height="200" fill="#FFFFFF" />
<image xlink:href="https://upload.wikimedia.org/wikipedia/commons/thumb/a/a6/Logo_NIKE.svg/200px-Logo_NIKE.svg.png" width="200" height="200" />
</svg>
So yeah, not bad.