3 ms·

Someone use GPT to automatically annotate articles with domain specific knowledge so that I can highlight words/sentences/symbols/formulae/etc. and have it expl

Someone use GPT to automatically annotate articles with domain specific knowledge so that I can highlight words/sentences/symbols/formulae/etc. and have it explain it to me.

bgirard 4y ago
There was a news site that did something like that, I don't recall which. It was incredibly distracting and I think they started annotating ads in their popovers as well.
- BobbyJo 4y ago
  This is why it's have to be a separate lyer from the page itself. Otherwise it would become yet more add real estate.
gtop3 4y ago
This feels like the type of thing I want human expertise on. GPT can be too inaccurate for something like this. I'd also be concerned it was trained on writings from people that misunderstand the science. There's already tons of expert written and reviewed content designed to teach science. Go buy any textbook on the subject. You can usually get them very cheaply if you don't mind older editions. Libraries will also have these types of books you can checkout for free. Once you've exhausted information found in textbooks you can start actually reading the scholarly journals these pop-sci articles are written about. Once you've read enough scholarly journals, you might have your own questions that aren't answered yet. Then, you can conduct your own science...... Books are great for looking up quick tidbits of information too. Chapter titles and indexes are great for jumping to relevant information.
- BobbyJo 4y ago
  I'm not a physicist, so keeping around a good deal of reading material (not to mention seeking out a good explanation) to understand the odd reference or two seems horribly cumbersome. Google is way lower friction, but even that requires some manual sorting through. Seems like exactly the thing a good language model would be great at. If they can't do that, where the stakes are low and the task is exactly the domain of LLMs, then I don't see how they'd be useful for anything.
  - gtop3 4y ago
    I'm optimizing for accuracy and depth of understanding. Your optimizing for lower friction. I think I'm optimizing for the correct variables.

Someone use GPT to automatically annotate articles with domain specific knowledge so that I can highlight words/sentences/symbols/formulae/etc. and have it explain it to me.

bgirard 4y ago

There was a news site that did something like that, I don't recall which. It was incredibly distracting and I think they started annotating ads in their popovers as well.

BobbyJo 4y ago
This is why it's have to be a separate lyer from the page itself. Otherwise it would become yet more add real estate.

gtop3 4y ago

This feels like the type of thing I want human expertise on. GPT can be too inaccurate for something like this. I'd also be concerned it was trained on writings from people that misunderstand the science. There's already tons of expert written and reviewed content designed to teach science. Go buy any textbook on the subject. You can usually get them very cheaply if you don't mind older editions. Libraries will also have these types of books you can checkout for free. Once you've exhausted information found in textbooks you can start actually reading the scholarly journals these pop-sci articles are written about. Once you've read enough scholarly journals, you might have your own questions that aren't answered yet. Then, you can conduct your own science...... Books are great for looking up quick tidbits of information too. Chapter titles and indexes are great for jumping to relevant information.

BobbyJo 4y ago
I'm not a physicist, so keeping around a good deal of reading material (not to mention seeking out a good explanation) to understand the odd reference or two seems horribly cumbersome. Google is way lower friction, but even that requires some manual sorting through. Seems like exactly the thing a good language model would be great at. If they can't do that, where the stakes are low and the task is exactly the domain of LLMs, then I don't see how they'd be useful for anything.
- gtop3 4y ago
  I'm optimizing for accuracy and depth of understanding. Your optimizing for lower friction. I think I'm optimizing for the correct variables.