5 ms·

Not sure why you put learning in quotes. It occasionally returns text reflecting true things that people asking question about don't know. Is that conspiracy th

by pizza 3y ago

Not sure why you put learning in quotes. It occasionally returns text reflecting true things that people asking question about don't know. Is that conspiracy theory?

akira2501 3y ago
I think there's a distinct difference between "learning" as is done in schools and "learning" as is meant by merely memorizing facts. I'm sure GPT can help you with the latter, I'm not sure it can help you with the former; particularly, because you have to know _what_ to ask before it returns anything.
- parhamn 3y ago
  I think i agree with the core of what youre saying but this certainly isn't true. Ask ChatGPT for a syllabus foe an introductory calculus course and ask for a study plan. Then ask for a study plan for each section. Than start asking it to explain things. People will build UIs to automate this process where all you have to enter is what you want to learn, in this case “introductory calculus”
- faeriechangling 3y ago
  Does school NOT consist largely of memorising facts? Do schools not consist of this more and more due to increasing reliance on metrics based on standardised testing, where the best way to juke the stats is to teach the test which means teaching kids to memorise facts? Ones which explicitly limit access to the internet or LLMs during test explicitly to make the only route to passing rote memorization of a great deal of facts? Does GPT and LLMs, which can easily crush any test based on fact memorisation, not present an opportunity to offload the task of mere fact memorisation to the machine and allow one to learn how to do things which would have previously been unviable without first memorising a great deal of facts? I don't consider knowing WHAT to ask to be all that much of an obstacle either, you just ask it about something you don't know very much about, but that you want to do. If anything I see LLMs as presenting an existential threat towards models of measuring merit through memorisation, as it's exposing said memorisation of being testing for testings sake, disconnected from real world utility or the ability to meaningfully measure merit.
  - akira2501 3y ago
    > Does school NOT consist largely of memorising facts? Is there a single standard for "school?" I was using it as a device to separate out the difference between the two styles. I do recognize that the majority of American schools are simply memorization factories but to focus on that is to miss the argument entirely.. > If anything I see LLMs as presenting an existential threat towards models of measuring merit through memorisation As you've essentially just repeated that same argument with this point. The types of "learning" LLMs are good at are not the types of "learning" people would _expect from a good school_. I do believe that this will occur, but it simultaneously undercuts your previous arguments in favor of GPTs utility, as they only make sense in a world where education is _purely_ about memorization. And sorry to reorder this, but... > which can easily crush any test based on fact memorisation, What is your source for this assertion?
reedciccio 3y ago
You can't trust what the LLM say. Occasionally true answers is the same as a broken clock: it's correctly displaying the exact time twice per day. Can't be trusted.
- ericlewis 3y ago
  And I had plenty of teachers that made me say: "Wait a second..." only to find out once I independently verified that they were full of crap. Can't be trusted.
  - gumballindie 3y ago
    So the plan is to replace low quality teachers with a low quality bot instead of improving education? Why does openai / chatgpt marketing sound so puerile?
    - Kuinox 3y ago
      If a teacher making a mistake is "low quality", no humans can be a teacher.
      - gumballindie 3y ago
        I should have worded it better - one mistake is not low quality indeed. What i had in my mind at the time is that if we do have some low quality teachers it means we should improve the system not adopt a known to be wrong tool. Back in the day when the internet was becoming popular there were loads of similar stories (likely openai’s marketing team is familiar with these) about people using the internet to learn. Recent times are proof that that was a mistake for many.
        Kuinox 3y ago
        > Recent times are proof that that was a mistake for many. You are creating your own conspiracy about how internet created conspirationist. Social Medias are the fault, not learning stuff on internet.
gumballindie 3y ago
Google and facebook are also occasionally returning true content. And thats the issue. If it was either right all the time or wrong all the time it would be less of a problem - although learning is a bit more complex than just mindlessly reading text. The risk is that people will assume that this software is right all the time and that means they will uncritically follow it.

by pizza 3y ago

Not sure why you put learning in quotes. It occasionally returns text reflecting true things that people asking question about don't know. Is that conspiracy theory?

akira2501 3y ago

I think there's a distinct difference between "learning" as is done in schools and "learning" as is meant by merely memorizing facts. I'm sure GPT can help you with the latter, I'm not sure it can help you with the former; particularly, because you have to know _what_ to ask before it returns anything.

parhamn 3y ago
I think i agree with the core of what youre saying but this certainly isn't true. Ask ChatGPT for a syllabus foe an introductory calculus course and ask for a study plan. Then ask for a study plan for each section. Than start asking it to explain things. People will build UIs to automate this process where all you have to enter is what you want to learn, in this case “introductory calculus”
faeriechangling 3y ago
Does school NOT consist largely of memorising facts? Do schools not consist of this more and more due to increasing reliance on metrics based on standardised testing, where the best way to juke the stats is to teach the test which means teaching kids to memorise facts? Ones which explicitly limit access to the internet or LLMs during test explicitly to make the only route to passing rote memorization of a great deal of facts? Does GPT and LLMs, which can easily crush any test based on fact memorisation, not present an opportunity to offload the task of mere fact memorisation to the machine and allow one to learn how to do things which would have previously been unviable without first memorising a great deal of facts? I don't consider knowing WHAT to ask to be all that much of an obstacle either, you just ask it about something you don't know very much about, but that you want to do. If anything I see LLMs as presenting an existential threat towards models of measuring merit through memorisation, as it's exposing said memorisation of being testing for testings sake, disconnected from real world utility or the ability to meaningfully measure merit.
- akira2501 3y ago
  > Does school NOT consist largely of memorising facts? Is there a single standard for "school?" I was using it as a device to separate out the difference between the two styles. I do recognize that the majority of American schools are simply memorization factories but to focus on that is to miss the argument entirely.. > If anything I see LLMs as presenting an existential threat towards models of measuring merit through memorisation As you've essentially just repeated that same argument with this point. The types of "learning" LLMs are good at are not the types of "learning" people would _expect from a good school_. I do believe that this will occur, but it simultaneously undercuts your previous arguments in favor of GPTs utility, as they only make sense in a world where education is _purely_ about memorization. And sorry to reorder this, but... > which can easily crush any test based on fact memorisation, What is your source for this assertion?

reedciccio 3y ago

You can't trust what the LLM say. Occasionally true answers is the same as a broken clock: it's correctly displaying the exact time twice per day. Can't be trusted.

ericlewis 3y ago
And I had plenty of teachers that made me say: "Wait a second..." only to find out once I independently verified that they were full of crap. Can't be trusted.
- gumballindie 3y ago
  So the plan is to replace low quality teachers with a low quality bot instead of improving education? Why does openai / chatgpt marketing sound so puerile?
  - Kuinox 3y ago
    If a teacher making a mistake is "low quality", no humans can be a teacher.
    - gumballindie 3y ago
      I should have worded it better - one mistake is not low quality indeed. What i had in my mind at the time is that if we do have some low quality teachers it means we should improve the system not adopt a known to be wrong tool. Back in the day when the internet was becoming popular there were loads of similar stories (likely openai’s marketing team is familiar with these) about people using the internet to learn. Recent times are proof that that was a mistake for many.
      - Kuinox 3y ago
        > Recent times are proof that that was a mistake for many. You are creating your own conspiracy about how internet created conspirationist. Social Medias are the fault, not learning stuff on internet.

gumballindie 3y ago

Google and facebook are also occasionally returning true content. And thats the issue. If it was either right all the time or wrong all the time it would be less of a problem - although learning is a bit more complex than just mindlessly reading text. The risk is that people will assume that this software is right all the time and that means they will uncritically follow it.