Are Large Language Models Really AI?

Editing posts to say random words is not deleting them.

Oh, you don't find that useful? If only there was a downvote button you could use to express your displeasure.

Worrying about that at this point is a waste of time, they got everything already if it's out there.

Now yes, but this was not long after ChatGPT went public and the entire AI hype bubble began. Once it got out that these AI companies were scraping anything and everything across the internet to "train" AIs without giving any credit, public pushback started.

when it comes to doing it on SO it is just performative, there is an edit history on the post, and it will just be reversed and you lose access to the ability to change posts going forward.

In addition to that, SO releases a quarterly Data Dump which is an offline copy of the posts which your edit wont be reflected in (if one was done between your post and edit), and if you ask me I think an AI company is going to grab the neatly packaged pile of data before scraping the site when the option is just sitting there.


but like if it makes you feel better, like your sticking it to the man, go for it.

Reddit really suffering from that kinda thing. There's so little organic content to begin with, seeing a thread where 1/3 of the posts have been censored because people were upset about porn or AI is depressing.

We just need to vibe code a bot to mass registers and upvote the wrong answer on stack overflow.

Longing, Rusted, Seventeen, Daybreak, Furnace, Nine, Benign, Homecoming, One, Freight car.

FFXI Strawberry Mango Forklift

I know editing SO makes little to no difference. Nuking my account as a result of edits as protest for user contributions being used for AI *is* the point.

Not sticking it to anyone, nor will it stop AI companies using it. It is simply a form of protest.

The fewer people contribute to So the less value it has. The more people purposefully get their accounts banned the worse for So. Again, this is the point.

Same goes for all public sites. No users, no data, no data for AI to be trained with.

Actual useful information, I just saw a video mentioning that something is drastically changing in the AI coding market.

https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/

I already had that plan on using Claude Code as planning and architecture, and Copilot to actual execution, given the request metric is more forgiving than the token metric from Claude Code.

BUT... Github Copilot put on hold new licenses, plus they removed most Opus models from any plan, leaving only Opus 4.7 on the super tier.

Idk, but I would bet the service is starting to get more expensive due to the increase of demand.

While my reply is off topic I have to say it.

Reddit has always been a cesspool. How can anyone ruin a cesspool like that?

In the long run, not having to pay actors and build movies from prompts, means junk like Fast & The Furious all the way to part 37845 can be made. The base will eat it up, they've shown they'll watch jingling keys for 90 minutes after the first one. lol

Repeat for various franchises

While slop appeals for a while, like real life slop, it tends to taste pretty awful once you've had it once or twice.

Relying on LLM to regurgitate old content in a new form is not a sustainable strategy.

https://www.reuters.com/world/white-house-accuses-china-industrial-scale-theft-ai-technology-ft-reports-2026-04-23/

This ***boils my blood a little. It's the whole rules for thee, not for me mentality.

Please don't steal the output of our content stealing machine. We stole that content fair and square!

I'd have believed this until I saw all the cases of Lawyers using ChatGPT to cite wrong case studies.

People will take mental shortcuts, and they sell this as "super intelligence" so...

All mushrooms are edible but some only once. - Terry Practchett.

The best part is you know that image was created by AI.

That's been my point since the very first post in this thread.

curious how much you guys are currently using AI on a daily basis when it isn't required by your job or something similar? I played with a few things like the song generation when it was posted here but within the last few months I haven't used it in any capacity beyond what google spits out when you search for something. Haven't bothered to look if you can block that from search results or not, though I imagine something exists by now.

Zero. I have long since developed my research skills to the point I can find information faster and more accurately then some algorithm that used reddit posts as it's source of truth.

I'm with Saeval on this one.

ChatGTP wasn't citing wrong case studies, it was making cases up and citing them.

The odd part is that AI has been known to do this for a few years now and lawyers using AI haven't bothered to look up the cases on legal databases that they already have subscriptions to just in case the citations are really hallucinations.

I use it to help with well defined tasks, like I mentioned earlier I used it to construct some JSON stuff I didn't want to do myself. It took me less time to do that than it would've manually, but I still had to edit it.

I've had it do some bash scripting here and there because I hate bash scripting and it gets me about 90% of the way there faster, with less effort. Similarly I've had some annoying command sequences I didn't wanna do (e.g. like escaping strings a certain way) I fed it to, but it maybe cut out 5 minutes from me just doing it on another box with python and copy/pasting.

I've done some code analysis with it, but even Claude Code needs a lot of handholding and most of the results are kinda mid or misinformed. It's not horrible but my rule of thumb with this is that 9/10 times, I'm faster finding what I am looking for rather than trying to get AI to do it for me and it's less error prone. I've only had luck running it against repositories I already know really well so I can ask very specific, pointed questions in a prompt, but even then...if I know the code it rarely gives me anything useful I couldn't do myself. If I don't know it, then I can't interpret the output anyway without learning it, so either way I have to spend time reading the code, might as well just find it in a way that helps me learn the code better.

The ability for it to practically interpret code and create output is not stellar. Using the JSON example above, if I fed it the repo and asked it to build a JSON file to hit a certain node, it would kinda get 90% of the way there.

I was writing a kernel module and I was trying to figure out if I needed to lock before reading from a certain struct, it told me no and pointed to the mailing list discussion saying it wasn't necessary. I looked at the code and it looked necessary but I figured I was wrong and trusted the AI. The kernel panics that followed were a good reminder to not trust that ***. The lock was necessary and I should've trusted my judgement

Maybe I'm just doing it wrong. Everyone thinks people like me are gonna be replaced by Mythos. Everyone reading that Mozilla blog post different ways to their own biases. OTOH for 20 years they've been saying that about every new fad, so we'll see.