When I say AI slop I mean extremely low effort where you ask ChatGPT to construct and argument for you rather than doing it yourself. Like the guy who couldn’t come up with arguments so he went and chatted about my pov with ChatGPT and posted it as proof he was correct.
A lot of fun can be done with AI like art. I am having a blast with Supapres songs and Bambideaths videos. But “Please explain this” and copy paste or “please help me come up with a reply to this forum post” is nothing we need here nor anywhere.
Even in complaining about it, one has to understand its limitations. AI only puts together what it found in its training data, according to the probabilities of what symbol follows what. AI slop is people slop.
Honestly tho, it’s just an extension of the confirmation bias that people already inately have, the internet boom itself had the same problem, and then when searchengines started doing their thing there it was again, and now with AI it’s the same thing all over again.
The core of the issue is that people (either by choice, by lack of understanding, or pure stupidity) seem to be dillusional enough to think that what an AI (Be it grok, chat jippidy or whatever) writes is “truth” in any shape way or form, lacking the understanding that “that” is not what AI does, it’s a glorified searchengine, of sorts, with the drawback that it does, and will, hallucinate. And the answers it provides are hevily algorithmically controlled by the training data it has been fed.
Personally it’s gotten as far that when someone tries to quote AI in an argument, I honestly stop listening to them because they’ve proven that they have no original, factual basis for their oppinion, it’s all based on the fallacy of AI accuracy which anyone with half a brain-stem knows by now that it isn’t RELIABLE information, all “facts” need to be double and tripple checked with authorative sources before you can actually use it for anything.
Now don’t get me wrong, I’m not against AI, but I am, however, against stupid people.
For those interested, here are a few valid, usefull ways to use AI:
Write an original article, with sources and references, ask AI to restructure it according to a well laid out plan (Be it formality, structural sanity or whatever)
Write an original prompt for a music/video generating AI complete with lyrics and other usefull information, and ask the AI to create a video/song/music for it.
Run ideas past it, asking it to rebuttal your point of view to see what common pieces of information you might be MISSING (do NOT use it to confirm your bias)
Use it as a search engine to find sites/sources of information about a subject (altho at this point, using duckduckgo might actually be faster)
TL;DR AI, when used CORRECTLY can be quite beneficial, but when used INCORRECTLY it not only becomes in many cases completely useless, but even dangerous.
As a short anecdote, me and a couple coworkers just the other day, asked jippidy what time it currently is in a couple places around the world that go with CET+?? time, which it answered well, then we asked for a third location which is on the other side i.e CET- and got an answer that was completely false (we are talking 5+ hours wrong), when the AI was questioned about it it admitted the mistake and said it was a “type-o”. THAT is how easy it is to get an AI to hallucinate and it will never correct itself unless challenged.
When using AI for feedback it is also important to set up rules for it. A lot of people think they are the next Jesus when they ask ChadGPT for feedback on their ideas and writing of which we have seen a few examples of in this forum (that guy that stopped arguing with me and started chatting with GPT instead and then later posted the convo convinced he was right for instance).
I don’t know how other common AIs such as Perplexity or Gemini works but ChatGPT is made to be your friend and encourage you. You can set up rules in it to get a better response when it isn’t worried about your feelings. As an example I asked about feedback on a piece I wrote for which it suggested small edits, I then put it in the project where I had set up rules and it basically told me to toss it.
Free versions are limited but people like free stuff. I would advice anyone who wants to use AI and learn how to use it to buy a subscription.
Base Instructions asside , Rules are abselutely key
For example i often start of with a Apex Reasoning Mode promt of sorts followed by something like :
"Red Team Examiner Mode.
Write 10 specific kill criteria for [type of work, e.g., “this [page or word count] strategy memo”].
Score my work 0–10 on each with evidence.
If total <85/80, write the most crushing rejection letter possible, then a precise salvage plan.
Here is the work: [paste work]"
You can ( should ) ask AI in Apex mode as i do all the time for various projects ( best research assistant by far ), on how to setup rules if you dont know how.
the rules setup thing doesnt quite work for me. or at least not consistently with gpt. like I tried to set it up to ask back questions if it doesnt understand instead of just writing a wall of text full of assumptions (scientific work). and sometimes its super noticible and itll ask 5 times for basic stuff, then the next fresh chat I open, it just doesnt use these rules at all. The rule for trying to argue against me instead of yes-sir-ing me doesnt ever trigger it seems. So I just ignore the fluff in the beginning and if it goes in the wrong direction, you gotta reprompt more forcefully.
but yeah, its very annoying when people JUST use ai. lots of papers are unreadable junk. on the other hand, lots of people use it to just rewrite their papers, which makes sense, we are scientists, not linguists or professional authors. but its really hard to tell which kind of paper it is, unless you read it. very annoying. and also, even if its used just to polish language, it sometimes still uses the wrong words or switches numbers around. And those types of errors are really hard to spot when reading thru. at least for me, maybe I just dont have the eye for this sort of thing.
Stuff that works nicely tho is that api version where you can manage exactly how to use the tokens, how to splice, where to look, how to structure the answer, scope of knowledge etc. its a lot of work to set up (was for me at least) but then it works like a charm. especially if you force it to quote things and then run a double back to look for said quote in a seperate “prompt”. I still dont trust it but at least I can have it link a clickable thing to the exact position in the doc. also doubles as a nice research assistant / paper sorting database
GPT takes rules more as suggestions, so enforcement is extremely weak [SIZE=1]( a known issue )[/SIZE] You would have to include the rules with literally every single promt, everytime or it waters down. [SIZE=1]( I dont and likely wont ever use it for now )[/SIZE]
Apex Reasoning Mode is a advanced ( overload ) input Grok mode, used for highly complex ( unusually complex ) problems with much greater accuracy. [SIZE=1]There are improved alternatives[/SIZE]
Only prob i have with Grok is that it often calls me “dude”. [SIZE=1]Yes i can fix it :)[/SIZE]
The thing i fear the most about AI is that it requires immense, yes immense amounts of power to operate. Power that could have helped humans directly instead.
All this energy waste because we need to ask an AI what the weather is outside lol, soon itll say its global heating weather today too, only 50 degrees celcius today.
