Ancestors

Toot

Written by J Λ M Ξ S on 2024-12-07 at 20:32

folks skeptical of AI, I am curious what you might think about the results if you ask one to generate a study schedule for learning something.

A somewhat reliable example is: build a study schedule for Real World Ocaml

Or simply: build a study schedule for X

Then follow up with: Create a list of resources I can turn to for the necessary information

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Descendants

Written by Jeff Forcier on 2024-12-07 at 20:45

@jmsdnns seems to fall under the "acceptable" category of "output you can/must vet manually in order to use"?

I might still worry about omissions (what if there's some really good resource a human would recommend that's missing) but that's not likely to be a deal breaker.

=> More informations about this toot | More toots from bitprophet@social.coop

Written by Jeff Forcier on 2024-12-07 at 20:49

@jmsdnns eg, I might (if I wasn't just generally leery of the entire category of tool due to ethics concerns) ask an LLM to point me at OCaml resources, but I /would not/ ask it a question like “what's the right way to do xyz in OCaml” because I'm unable to verify that it tells me the truth. Could easily be repeating something old/wrong or just making shit up.

=> More informations about this toot | More toots from bitprophet@social.coop

Written by J Λ M Ξ S on 2024-12-07 at 21:01

@bitprophet how would you feel about: what libraries do people usually use for xyz in Ocaml

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Written by 0xC0DEC0DE07E8 on 2024-12-07 at 21:07

@jmsdnns @bitprophet aren’t these things for which there are fairly easily searchable answers for that don’t consume gobs of energy for each identical query?

=> More informations about this toot | More toots from c0dec0dec0de@hachyderm.io

Written by Jeff Forcier on 2024-12-07 at 21:08

@c0dec0dec0de @jmsdnns this is the other big reason I eschew the category, ya.

=> More informations about this toot | More toots from bitprophet@social.coop

Written by J Λ M Ξ S on 2024-12-07 at 21:09

@bitprophet @c0dec0dec0de im not trying to say we should use it. i am asking for opinions about the results.

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Written by Das on 2024-12-23 at 04:19

@jmsdnns @bitprophet @c0dec0dec0de opinion but disregarding cost (and possibly ethics)? It's not a hypothetical question as these queries can be actually run (and many are running them) - so what's the cost of the query? (vs other methods - is that the 'convenience' fee)

=> More informations about this toot | More toots from SRDas@mastodon.online

Written by J Λ M Ξ S on 2024-12-23 at 04:27

@SRDas @bitprophet @c0dec0dec0de i only expected a few of my friends to actually do it

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Written by Das on 2024-12-23 at 04:55

@jmsdnns @bitprophet @c0dec0dec0de this is the dilemma no? There are cases/scenarios where it could be useful. And this is an active area of research too.

(So I'm kind of wondering the same as you but wary too)

=> More informations about this toot | More toots from SRDas@mastodon.online

Written by J Λ M Ξ S on 2024-12-23 at 05:41

@SRDas @bitprophet @c0dec0dec0de that's right! one of the big concerns for people putting money into LLM based tools is that users dont seem to care about them. cant trust the nondeterministic nature of the outputs. what's the definition of insanity, right?

more generally, dr lilach mollick found AI is beneficial for lower performers in some context. AI boosts them up a lot. higher performers dont get the same boost and tend to think it isnt worth using

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Written by J Λ M Ξ S on 2024-12-23 at 05:58

@SRDas @bitprophet @c0dec0dec0de as i thought on that, i started wondering about things some people struggle with, like planning long term tasks. some of my friends are so bad at this.

that's what lead to wondering how well an AI might plan a study schedule for someone. wouldnt even need to use the AI again after that; just try to follow the schedule

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Written by Jeff Forcier on 2024-12-07 at 21:09

@jmsdnns this one seems similar, you can verify whether what it spits out is what you wanted, but you will also have to wonder if it's lying through omission

=> More informations about this toot | More toots from bitprophet@social.coop

Written by J Λ M Ξ S on 2024-12-07 at 21:10

@bitprophet there was an interesting example not long ago where an LLM kept hallucinating the same library, so some people created an actual library with that name and used it for malware

=> More informations about this toot | More toots from jmsdnns@mastodon.social

Written by Jeff Forcier on 2024-12-07 at 21:24

@jmsdnns could you please 2020s a little louder for those in the back? 🤣

=> More informations about this toot | More toots from bitprophet@social.coop