I'm doing an academic lit review on a topic and chatgpt just created a table for me on most cited papers in top 20 journals in the sub-field.
😭
its imperfect, but also amazing help.
@Beanc
Just make sure they all exist, and that you can find support for your thesis! I've been skimming a LOT of literature lately for my own projects, and I'm amazed how well keyword searches in PDFs facilitate homing in on what I want.
@Helical_Code
Always do.
Requiring source links helps it not lie 😁
@Beanc @Helical_Code I always tell these things to cite their sources; and they cheerfully fabricate some. Some are getting better about it.
The next level of problematicness* is that we can't know what the LLM was fed, and what is wasn't. So, what biases does that introduce? Don't know!
*a word I totally didn't make up
@b4cks4w @Helical_Code
I guess the answer is you never rely solely on the results, you rebuild history based on forward and back citations.
@b4cks4w @Helical_Code All that said, there's a dude I konw that is working to create a research AI.
probably subject to the same limitations as its base, I think, is chatgpt, but it is super helpful.
beta stage, rn, I believe.
@Beanc @Helical_Code You might know two then ;-).
I'm not using LLM per se, am just putting a front end on a static-ish corpus. Just to have a better nature language query tool. I see the potential.
@b4cks4w Small world =D
Haha let me know when you've got a beta I'll test it for you
@Beanc @Helical_Code No silver bullets but I remind myself how much better faster this is than wandering the stacks. Which I loved doing when they were a thing. Okay grandpa, time for dinner.
More efficient, at least.
I mean, 20 years ago, I could probably have a pretty good handle on most of the key areas within my area's body of literature in my field, but today????? Not a chance.
So many branches of the tree.
@b4cks4w @Helical_Code
a) require links. Then it's an easy check.
b) Sure, though I limit it to X journals and ask for top X by citation count and require X keyword or title hits.
Of course I have no idea if the outpput actually meets my paramters (other than keywords), but frankly, if you have a list of 20 as a starting poitn, you basically back into the entire literature fairly easily.
I wouldn't use it as my sole source, but man its helpful if you're digging into a new sub-topic.
@Beanc @Helical_Code ...which is also a problem with Google Scholar. Nice tool but the index could be changing daily, no idea what's in it.