Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPT makes learning fun again

vipshek.com

Post reply on HN
Print view View original

171–180 of 243 posts

Re: GPT makes learning fun again

#171

Post by notamy »

lcuff wrote: post #25
devjab wrote: post #7

I'm in two minds about it. On one hand the internet (in general) has become so hard to find information on, and I agree with the author that GPT is a breath of fresh air. On the other hand I've seen GPT fail so miserably at topics that I'm knowledge about that I have a very hard time trusting anything it tells me. I'm not sure what the answer really is, but I'm not sure it's GPT. I wish we could go back to having sea…

I'm curious about what topics you have knowledge of where it's failed. Does it seem like there's a pattern to the failures? I've been using GPT for coding help, and it it is very helpful in ruby and bash, though it often delivers buggy software: Badly handled non-happy-path conditions, mostly, which when I tell it to handle the case, it may. It's a huge help for me finding gems and showing me standard ruby library sy…

> Does it seem like there's a pattern to the failures?

Personally, trying to use it to write code (primarily Elixir backend and Rust systems/CLI), I tend to run into:

- Hallucinating APIs that don't exist

- Hallucinating entire libraries, despite being told repeatedly they don't exist

- Saying it will make requested changes and not doing so

- Not being anywhere close to idiomatic code

- Not being able to explain code it writes

- Running out of "memory" (I can't remember the right term. Context?) in the middle of generating code, then telling me I never prompted it when I ask it to continue

On the other hand, I've found that it's good at cleaning up ugly data. I can copy/paste in a table with bad formatting, ask it to turn it into code, and it does it near-perfectly. That's been the best use-case for it I've found so far.

I use boring normal free ChatGPT so maybe it's on me for not using GPT-4 or some other model, but either way, imo it's not been very impressive in the problem spaces I find myself in.

Re: GPT makes learning fun again

#172

Post by burnished »

ChatGTP wrote: post #142
rolisz wrote: post #105

Earlier quoted context omitted.

I have a finite amount of time. I can choose to learn and know more about the software that I'm developing and make the software better, or I can spend some amount of time to learn the DevOps part, or some mix in between. In some cases, the ratio is 90:10 in favor of software development.

Is nginx hard to understand or something ? in surprised how few services developers I’ve encountered who don’t know hot to configure a web sever or LB.

Yeah honestly when I took a stab at it a couple years ago it was frustrating - all the material I ran into told you what to do but not why which left me high and dry when things didnt work as expected, or when I wanted to find more information about the meaning and organization of directories or other contextual background information.

Re: GPT makes learning fun again

#173

Post by mrtranscendence »

dividedbyzero wrote: post #125
blibble wrote: post #106

Earlier quoted context omitted.

why have no ads when you can plaster it with ads and also charge $20? see: windows 11

Would LLMs even have to be plastered with ads? I suspect just being subtly biased would go a very long way and be very hard to detect.

Introducing subtle bias for any advertiser that buys ads might be a tough nut to crack. For one thing, advertisers and the companies that employ them tend to want creative control over their ads. And it's not clear that a subtle bias towards Palmolive liquid detergent would even do anything in a discussion about veterinary care for ferrets (or whatever), so would the advertiser have to pay for that? Or would they only pay if the user clicked on a link? But if the user has to click on a link, then it's no longer subtle bias.

Edit: Perhaps Colgate-Palmolive could advertise Palmolive on ChatGPT by having it so that, anytime liquid dish detergent does come up naturally in conversation, then the chatbot subtly points to Palmolive brand. But success here would be hard to measure.

Re: GPT makes learning fun again

#174

Post by dmd »

hammyhavoc wrote: post #65
LASR wrote: post #64

Earlier quoted context omitted.

I want to use this comment to state one thing, not directly addressed to you. Stop using GPT as a database! GPT is far more useful a reasoning engine that can accumulate fuzzy data and then provide various views or transformation of that data. So asking GPT to parse a Wikipedia page and then asking it to teach you from it - this is a much more successful usage than what the author in the original article is doing. It…

Is it? I couldn't even get it to figure out an NGINX config, despite feeding it the documentation and URL rewrites it would need, and the prior Apache .htaccess file. Spent days on it. Consulted with people who know more than me about prompt engineering. Nope. Reasoning is not its strong point, IMO. It's a next-word prediction model, why would it be? It's doing what an LLM does. Frequently, nonsensically.

> It's a next-word prediction model

So are you. https://ceoln.wordpress.com/2023/03/31/its-just-predicting-t...

Re: GPT makes learning fun again

#175

Post by Capricorn2481 »

ziml77 wrote: post #160
throwaway2990 wrote: post #152

Earlier quoted context omitted.

Except ChatGPT often gives better more accurate answers than Google without the bias of a single web page.

Instead you get arbitrary bias without multiple sources you can check. If there is a shitty source that comes up in a normal web search, I can skip it and go to the ones I know are better. How do I do something similar with ChatGPT? Am I supposed to ask it multiple times to average the randomness that's applied to the conversation? Is the average even what I want? There's a ton of garbage information on the internet.…

Same. Frequently the context gets screwed up and I need to start a new conversation because ChatGPT gets so far up its bum

Re: GPT makes learning fun again

#176

Post by devjab »

lcuff wrote: post #25
devjab wrote: post #7

I'm in two minds about it. On one hand the internet (in general) has become so hard to find information on, and I agree with the author that GPT is a breath of fresh air. On the other hand I've seen GPT fail so miserably at topics that I'm knowledge about that I have a very hard time trusting anything it tells me. I'm not sure what the answer really is, but I'm not sure it's GPT. I wish we could go back to having sea…

I'm curious about what topics you have knowledge of where it's failed. Does it seem like there's a pattern to the failures? I've been using GPT for coding help, and it it is very helpful in ruby and bash, though it often delivers buggy software: Badly handled non-happy-path conditions, mostly, which when I tell it to handle the case, it may. It's a huge help for me finding gems and showing me standard ruby library sy…

I can give you a few examples.

Some of the “issues” are more general, as in it giving you dated answers. I asked it so build some ODATA things in Typescript and C#, and it did to varying degrees of success, but some of the code was deprecated. Some ranging to “you should never, ever, do things this way” to “well IActionResult was replaced by ActionResult but it hardly matters”.

Others was where it made things up once pressed upon being wrong. I think the two most hilarious situations was when it did its “Sorry, you’re right…” thing and then proceed to give the exact same answer it had just given prior. The other was when it made up a function that had never existed, it has a very convincing name and at first I thought it was again a matter of something deprecated, but it turned out to be completely made up code. Some of this is down to me not prompting it right, but part of it is also worrying. Because what really made me worry about it was when I joked with it. I read (listen to) a lot of Warhammer audiobooks from Black Library, and, when I jokingly asked it something silly about Khorne at one point, it led me down a rabbit hole of discussing books with it. Books it obviously had never read, but would still confidently tell you about wrongly. Maybe it got its knowledge from the internet, maybe it made it up, but what was interesting to me about it was that if it would be so confidently incorrect about those books, then what else would it be confidently incorrect about.

This isn’t just a GPT problem of course. If you want to learn something, you need to consider the sources you use. When I went to folkeskolen (school for children aged 6-14) we were taught the pyramids were build by slaves. Later that has been disputed because of how well fed the workers were, but if you wanted to learn about the pyramids and you read my old school books then you wouldn’t get the updated knowledge. Similarly a lot of the sources you can find for learning programming are outright terrible, but outside of GPT you tend to be presented with a myriad of choice to remind you that some sources are better than others. With GPT the source or what it teaches you isn’t obvious. If you wanted to learn about the pyramids, you probably wouldn’t pick a 30 year old school book for children after all.

Re: GPT makes learning fun again

#177

Post by throwaway2990 »

ziml77 wrote: post #160
throwaway2990 wrote: post #152

Earlier quoted context omitted.

Except ChatGPT often gives better more accurate answers than Google without the bias of a single web page.

Instead you get arbitrary bias without multiple sources you can check. If there is a shitty source that comes up in a normal web search, I can skip it and go to the ones I know are better. How do I do something similar with ChatGPT? Am I supposed to ask it multiple times to average the randomness that's applied to the conversation? Is the average even what I want? There's a ton of garbage information on the internet.…

If 100 people write about a topic. And 2 of those people are wrong. And you searched and for some reason the SE ranked one of those 2 higher. How will you know or spot incorrect information or bias?

If the information on ChatGPT is averaged out the chance of it being correct is high.

If you ask questions with bias in them you can sway the results of ChatGPT. That just means to me you’re asking the wrong questions.

In terms of devops. Programming. Configuration. Daily tasks. Generalised workflows. How-tos. Etc. ChatGPT will more than likely give you accurate results that are useful, or pretty damn close.

Obviously with information being limited to 2021 then it can be hard to find some solutions. I needed to do something with named pipes in c# that I couldn’t get working in .net 6. The examples it kept giving me were for .net core 3.1, or .net framework. When I asked specifically about .net 5 (since it doesn’t know .net 6 is released or that 7 exists) it apologised and said the code won’t work in 5 and gave me a working example in .net 5. Which identified that there was an api change I wasn’t aware of.

No amount of googling found a solution for me.

Re: GPT makes learning fun again

#178

Post by Capricorn2481 »

p-e-w wrote: post #22
pavlov wrote: post #12

Earlier quoted context omitted.

Kindness and patience have always been in short supply on the public internet, but AI can simulate them in infinite amounts. That’s a positive thing about this generative AI revolution that I haven’t really thought about in those terms until now.

That is actually the killer feature of interactive AIs. People go on and on and on about "accuracy", completely ignoring that accuracy is irrelevant to 99.9% of things that humans do in their everyday lives. Simulating (positive) human interaction is far more impactful than getting facts correct.

> Simulating (positive) human interaction is far more impactful than getting facts correct.

In my experience, people with the attitude that AI will give them a more fulfilling relationship than humans will aren't exactly friendly to begin with.

Re: GPT makes learning fun again

#179

Post by burnished »

thefz wrote: post #72
simonw wrote: post #24

Earlier quoted context omitted.

"I'm not sure what the answer really is, but I'm not sure it's GPT." My suggestion is to stick with it and get a feel for what it's good at. I've found that after a few months of using ChatGPT every day I've developed a pretty solid intuition for which questions are likely to get good answers and which are likely to trigger hallucinations. It's difficult to describe what those intuitions are though! One rule of thumb…

> One rule of thumb I've developed: if something is likely to be "common knowledge" - if it's something that is likely to have been discussed accurately on the internet by many different people - then ChatGPT is very likely to answer questions about it accurately. If so, this information is already easy to find, making GPT redundant.

Let me give you a concrete counter example; I wanted to review DNS records the other day which is a topic I had covered in undergrad. Searching on google retrieved a bevy of poorly written articles that were mostly noteworthy for bow they glossed over the material I wanted to review.

I asked chat gpt to review the topic for me and it wrote a helpful summary that tracked with what I remembered and provided enough keywords that provided better search results - fact checking and digging for more detail became much easier when I was able to find the exact wikipedia page I needed (and other resources).

Re: GPT makes learning fun again

#180

Post by brokencode »

phh wrote: post #91
brokencode wrote: post #19

I noticed a while back that the internet has made me terribly prone to skimming. It got to a point where I could hardly stand to read anything longer than a short news article. To learn anything useful on the internet, you pretty much have to skim. So much of the internet is so loaded with filler and BS that it is hardly worth reading at all. With ChatGPT, it’s incredibly refreshing to be able to ask a question and g…

It's interesting because that's my biggest gripe against ChatGPT, it's too verbose and very hard to skim through. Compared to stack overflow answers it'll be easily four times longer. Sure compared to for instance news website, it is less verbose, but those are easy to skim. I'm still able to read through well written text without skimming, though yes most of the texts I'll read I'll skim through, then go back and th…

I think it depends on the subject, as this isn’t something I’ve had issues with personally. I’d suggest asking ChatGPT to respond with concise answers. Or perhaps tell it that you are familiar with the subject and don’t need the basic details explained.
Post reply on HN