User talk:TurboSuperA+

Archives

[CTOPs] • [1] • [2] • [3]

Active talk page discussions

AI tag at Lando Norris

Hiya, just curious which sections are under question of being written by an LLM at Lando Norris, so I can work on fixing any potential issues. MB2437 21:32, 14 July 2025 (UTC)

It is the Lando Norris#IMSA section. The whole paragraph has a single source and the link contains "?utm_source=chatgpt.com". TurboSuperA+_(connect) 06:04, 15 July 2025 (UTC)

Thank you, I will work on rectifying this. MB2437 06:28, 15 July 2025 (UTC)

Thank you, and sorry for tagging the whole article. I did it because there was no template for sections, but there is now: Template:AI-section. TurboSuperA+_(connect) 06:35, 15 July 2025 (UTC)

No worries! I am concerned there may be a wider issue now so I will scan through the rest—a lot of the article needs work anyway. MB2437 06:43, 15 July 2025 (UTC)

I found the article (and the section) by searching for insource:"chatgpt.com". Editors who use ChatGPT are too lazy to even remove that bit from the link, they just copy-paste the text without thinking or reading it over. TurboSuperA+_(connect) 06:45, 15 July 2025 (UTC)

Wikimedia may need to crack down on it, given it is textbook plagiarism. MB2437 06:54, 15 July 2025 (UTC)

The worst thing about it is that LLMs are often wrong, but they "hallucinate" sources so at a glance it all looks well-sourced. TurboSuperA+_(connect) 07:14, 15 July 2025 (UTC)

They can't even get book quotes right; I asked GPT for a list and it hallucinated 20 of them. I told it to stop making things up and it immediately did so again! It's worrying that people can hastily add such information without verifying what's been said. I find LLMs very useful for idea generation and some code fixes but nothing more. Fortunately, this instance only got his class position and total laps completed wrong; at least the overall positions and other details were correct. MB2437 07:29, 15 July 2025 (UTC)

I told it to stop making things up and it immediately did so again!

When I use LLMs it feels like I am correcting them half the time. This "AI" thing is nothing but a fad, a pump-and-dump scheme. Some of these companies have valuations in the billions of dollars. It's like we haven't learned anything from the dotcom bubble, or any of the other bubbles.

I find LLMs very useful for idea generation and some code fixes but nothing more.

It's basically a more advanced Google search. I think a lot of them have code execution now, so one can debug errors immediately. When it comes to generated text, a person has to "debug" it "manually".

this instance only got his class position and total laps completed wrong;

It always gets something wrong. The editor who added that paragraph has been adding lots of LLM-generated text and sources to Wikipedia. I left a comment on their talk page asking them to stop using LLM to edit. They haven't been active this month, so I hope when they come back they agree to not use it any more. TurboSuperA+_(connect) 07:53, 15 July 2025 (UTC)

I believe it will become of good use eventually, but we are decades away from that point. I find I'm spending more time verifying its "work" than I would be if I did the research myself. It was completely useless when I wrote my engineering thesis barring its use in programming—admittedly, that was more to do with me being useless with MATLAB syntax.

I'm surprised this instance slipped under my radar given they had completely removed an entire unrelated section to add it. I remember seeing someone correcting the section's heading and being confused (given I wrote the headings) but at least I now know I wasn't being a fool after all. MB2437 08:24, 15 July 2025 (UTC)

I believe it will become of good use eventually

Maybe. Right now they're using it to spy on people, target ads and make autonomous killing machines. I'm not going to hold my breath for AI innovations that will improve the life of regular people. If AI could do laundry, tidy the home, clean the dishes, clean the floors, every single household would have an AI. As it stands now, they're using it so the toaster and the refridgerator can collect data on you.

I find I'm spending more time verifying its "work" than I would be if I did the research myself.

Yup. Honestly, it's nothing more than a curiosity. I mostly use it for things like getting a one-liner ffmpeg command to convert a video and keep it under a certain file size.

barring its use in programming—admittedly, that was more to do with me being useless with MATLAB syntax.

I wish LLM was around when I wrote my thesis (also involved MATLAB).

I remember seeing someone correcting the section's heading and being confused (given I wrote the headings) but at least I now know I wasn't being a fool after all.

One has to trust one's hunches. They exist for a reason! TurboSuperA+_(connect) 08:34, 15 July 2025 (UTC)

Clean the floors—Roomba has been doing it since 2002! It will always take a bit of engineering genius to solve those issues. With defence budgets ballooning, there is no doubt such genius will be misused. I wish LLM was around when I wrote my thesis—honestly, it saved me from losing my mind on a fair few occasions. It's not the best with wikitext, unfortunately, but was able to solve a few template issues I've had. Mostly on me for sticking to my guns and refusing to use Lua. One has to trust one's hunches. They exist for a reason! I have no clue why I didn't investigate further, low-IQ moment on my end; must've been busy! MB2437 08:48, 15 July 2025 (UTC)

Roomba has been doing it since 2002!

Poorly. Roombas just vacuum a bit and they cannot get the corners, under the sofa, etc. all the places where the dust gathers.

refusing to use Lua.

Lua is easy. Sure it is weird, and most of the things you learn aren't transferrable to other languages, but it does have its own logic and it is easy to learn.

I have no clue why I didn't investigate further

Because of WP:AGF and trust in people, probably. TurboSuperA+ 08:29, 16 July 2025 (UTC)