• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Share your 'arguments' with ChatGPT or other bots

Safe-Keeper

My avatar is not a Drumpf hat
Safe-Keeper
Joined
Jun 18, 2007
Messages
13,774
Location
Norway
I'm asking ChatGPT for help with coding for the game I'm working on (/shameless self-promotion), and I keep getting an error message for a script it gave me, which seems to be caused by a certain bit of code appearing twice in the script. Problem is, no matter what I tell it, ChatGPT insists that 'word' only appears once. It'll even give me new scripts that are supposed to work, but where that same bit of code still appears repeatedly.

Anyone else had some weird disagreements with ChatGPT where, like some inescure dude in real life, it just refuses to admit it's wrong?
 
It's since been fixed, but my favorite example of this is:

ME: "How many Rs are there in the word strawberry?"
CHATGPT: "There are two Rs in the word strawberry."

I'll never fully trust anything any AI tells me.
 
Yes, the occasions when they derp out and get stuff outright wrong are bizarre.

Claude: * * * *
Me: Why are there only four lights here, not five?
Claude: You're absolutely right! There should be five lights, but there are only four.
Claude: Let me look into that.
Claude: I have corrected the code to properly display five lights.
Claude: * * * *
Me: No you didn't, there's still four.
Claude: You're absolutely right! [repeat ad nauseum]
 
Well, I just tried the "How many R's are in the word strawberry?" thing and it gave me the correct answer, fwiw.
 
An AI doesn't think. In essence it's a huge complicated pattern matching machine. It simulates text that it has learnt from reading millions of texts from the Internet and fills in combinations of the most likely words. It has some ability to parse your questions, but you have to remember that it can't really think or reason like we do (yet?). It doesn't really count letters, it just gives a more or less likely answer based on what it has seen in other texts before.

Management at work insists we should use more AI to save time, but the only valid use case I can see is searching through our documentation and giving the right answers faster. However my experiences with Google Gemini AI during Internet searches isn't exactly stellar. Initially you get a nice quick summary, but when digging deeper it turns out that some answers were wrong in a rather subtle way. At that point I had spent already some time working with that wrong answer, so in practice it took a lot more time than first digging through docs myself, puzzling together a right answer immediately. That means that this use case is also dead in the water to me. When I search for information I don't really need a quick answer, but I absolutely need a right answer!
 
Don't use it for search, not even google can manage that. Think of it like an over-eager but naive assistant. Great at grunt work and making first drafts for you to polish, but don't trust them implicitly. The most solid use case is formatting boilerplate responses. "Here's my spending justification for 2023, use it as a template for these 2025 receipts" sort of thing. Or if you code, they'll get you over the "ughhh I need to make a web scraper for this, how in the absolute **** does BeautifulSoup syntax work again" hurdle. Either way, what they give you is a first draft - you'll still need to review it. Don't give it to a judge or try to administer the US government without strict observation.
 
Beelzebuddy said:
Don't use it for search, not even google can manage that. Think of it like an over-eager but naive assistant. Great at grunt work and making first drafts for you to polish, but don't trust them implicitly. The most solid use case is formatting boilerplate responses. "Here's my spending justification for 2023, use it as a template for these 2025 receipts" sort of thing. Or if you code, they'll get you over the "ughhh I need to make a web scraper for this, how in the absolute **** does BeautifulSoup syntax work again" hurdle. Either way, what they give you is a first draft - you'll still need to review it. Don't give it to a judge or try to administer the US government without strict observation.
Click to expand...
Yeah, having just gone through three years of uni, and having started just as AI became mainstream (the same way I entered middle school just as the Internet became commonplace), I like to think of AI as something you use the same way you use Wikipedia: A great starting point that gives you enough understanding to take on the actual assigned reading with less difficulty, but nothing you should use as an actual source. Also it's great at giving you feedback on your writing, shortening or simplifying lengthy texts, etc. I typically don't use the end result itself (especially not when writing papers!), but I take whatever I agree with and learn from it.
 
I finally managed to convince ChatGPT that there were two lines of code, but it still blamed me for the mistake, implying I had accidentally added something to the code it wrote and gave me 🙃. Oh well, progress is progress, I guess😅.
 
You must log in or register to reply here.

Back
Top Bottom