Yesterday or last night to be specific, I asked Claude AI to perform a small task that involves scraping a website. It took him some revisions to achieve the way I expected. I had one another website to scrape, but I ran over the limit. Claude by-default give you long answers, so it’s pretty easy to run out of limit with him. Then you have two options left, either switch your account, or switch the AI. I am a person who likes to keep everything in under one account, so it was not valuable for me to switch accounts. Instead, I find it easier to switch GPTs. I rarely ran out of limit from ChatGPT, so I switched to him. And I found that, it took way too many attempts to achieve the same task. This was surprising because the task was pretty basic for an AI.
What was my Query?
Before I start my rant with inability of GPT to produce a simple task. I’d like you to know what I was aiming for. I was an active participant of an English Learning server on Discord. Where I usually post something that helped me improving my own skills rather than some copy-pasta that I acquired from one of the 1000s websites or videos. I had a great idea to scrape “The Office” quotes and highlight the prepositions used in those quotes along with their purpose or usage. For example, ‘”The worst thing about prison was the dementors.” – topic‘, this clearly highlights the preposition used here is about and explain the purpose for which it was used here, the purpose here was a topic. Here is another example, ‘”There’s a lot of beauty in ordinary things.” – part of whole‘, clearly highlights the preposition and its purpose.
How it goes with Claude AI?
Though it was simple for Claude AI to make a list, it still took me multiple revision, five revisions to be specific, to achieve what I was looking for. Off course it refuses to provide me the list straight away the first time I asked it for because he assumed those quotes were copyrighted.
Since when they send you a Copyright strike either on YouTube or in general through traditional sources on using their quotes? Anyway, I made a case and then he proceeded with it.
Then the first few revisions had incorrect assumptions of preposition which filtered out in the final revision, I was happy with the result.
I was thinking for doing the same with Brooklyn 99, but I wasn’t able to do that because I ran out of limit. I decided to move to Chat GPT with the clear picture of what I wanted, with a website that have 80 quotes from that series. Everything was set already, all I had to do is to put the list, ask GPT to scrape it and I would be ready in a minute. Oh boy! I was wrong! I was so wrong! It was a train-wreck that took forever until it delivered it.
ChatGPT A Million Dollar Train-Wreck of a GPT
I send a screenshot of what I had achieved from Claude to ChatGPT, expecting him to follow the same format. ChatGPT assured me that he can do the same once I provide him the link.
I provided him the link and then he gave me a sample on how he should approach the process. It was looking promising in the beginning.
But then he said, he couldn’t be able to fetch the link. It wasn’t something unexpected, it happened a lot with me with ChatGPT. I copy-pasted the code in GPT prompt to make it easier for him to work on.
Now, instead of giving me the result straight away, he gave me another sample format. In my very first request, I gave him the screenshot of the format I was looking for. Where did the “Explanation” come from?
Anyway, I asked him to generate the result whichever way he wanted to, and this was the result.
This was fine, but it includes the explanations, and he was also producing in batches. In fact, he only provided me the first batch, which was weird. I asked him to remove the explanations, he removed the usage as well.
For some reason, he forgot that I provided him copy-pasta already.
Finally! Oh finally! I got the list I should’ve got earlier. After couple of prompts. Was the secret sauce of working with AIs is to mention their competitor’s name? Who knows? I mentioned Claude made a list easily and it’s you who’s having such problems.
But there was a catch, it wasn’t complete. And even after couple of prompts, it didn’t produce the complete list. Until I decided to make a new chat and asked the same again. Is ChatGPT getting dumber in 2025? Well, it seems so.
Why use AI when you can Copy-Paste yourself bro?
Now you must be wondering something, why I didn’t do it myself? Why it was too hard for me to highlight preposition going through every single quote, around 200 in total, and then identify its usage; the link I shared with Claude AI for “The Office” had 120 quotes and the link I shared with ChatGPT had 80 quotes from “Brooklyn 99”. My purpose was to paste them in Anki for space learning. So why couldn’t I do everything manually?
Because it’s a waste of time! And in fact, I still able to save my time doing so with Claude AI. Off course I wasted a lot of time doing so with ChatGPT already.
Assessments
It is not clear which version of ChatGPT model you must be using, and it only become clear when you ran out of usage limit. I am assuming I had limited access to GPT-4o remaining on my account at that particular moment. I asked ChatGPT today, what does model=auto parameter means in the URL? Considering it was my first chat of the day. He told me, it is typically 3.5, and as a paid user I will get access to GPT-4-model, typically GPT-4o. This is contradictory to their price table and my experience as well; price-table clearly shows that I will get limited access to GPT-4o as well as I did mention that I had received notifications when my access to 4o ran out. Claude clearly shows the version you are using, Version 5 in my case. In comparison to ChatGPT which for some reason obscuring the model version in use which is pretty strange to me.
Now I understand that AI is still in its early age, but I don’t expect that level of mistakes from GPT. In fact, it was having a hard time to keep something in memory in an active thread. Full disclosure: my chat memory was full, but it was full for a long time, and I don’t mind clearing it once I really need to be. ChatGPT do remembers the active-thread conversation, most of the time. He does forget sometimes. Is it because ChatGPT is getting dumber? Look, I am not sure here. Maybe it’s getting too human that it catches human behavior to forget in mid-conversation.
Am I the wrong here comparing two different GPTs using two different Models?
Before you point me out that I was comparing two different LLM models for two different GPTs as a free user. Where ChatGPT gives you unlimited (with fair usage policy I believe) access to chat, on the other hand, Claude AI chats are very limited. Remember for a non-programming purpose, I don’t need AI beyond a point. I could use Gemini for programming and Claude for reasoning, and GPT for my daily driver. I may not consider ChatGPT as a paid option in future. In fact, I may not consider any of them to be paid unless I have any industrial requirement attached with me. So, even on the face value, my assessment stands strong.
Anyway, you are free to put your opinions or challenge my views or observations. I will be glad to know your thoughts or experiences with ChatGPT or any other GPT. Please feel free to use the comment section and thanks for reading this article, I will publish a new article soon. Have a great day!
