Can we, please, not fill posts here with slop? Detector say: 50% AI or likely AI and given that it's exactly what it says about typical ChatGPT Pro generated texts I'm not sure if there are anything human-written.
The whole post may be written as three or four lines of text without making people read the slop…
Why should we care whether the post was human-written? I don't see any quality issues with the way it was written.
Whether the benchmarks are of interest is a different question.
We don't. But we do care about netector marks. Detector doesn't really detect AI-written texts. It detects texts that are endlessly repetitive and contain very little information in the piece of text.
Whether you are creating such slop using AI or using some other means is irrelevant: AI creates such texts because it's not penalized and, in fact, often incentivized to write long texts with very little content (while human have disincentive: human time is limited), but one may force human to write something like that, too.
When someone would teach AI to write short and concise texts (and these would, ievitably, score low marks on detector) — I would be happy to read these.
Because if you wanted to talk to an LLM, couldn’t you sign up on the provider’s website and use it that way?
Another way of putting it is: I think most people are more interested in seeing your prompt than your output.
The OP does not at all appear to me repetitive or bloated with little information. So I have no idea what the problem is, or what that detector measures or why we should care what that detector says.
Modern AI like ChatGPT is in fact short and concise if you tell it to. I am very happy with the answers I get.
That's like saying: why do we need people in the forum, if you wanted to talk to people, couldn't you go somewhere where you can meet people?
The OP had some information to share.
It also detects them with... AI. Which is more concerned about matching similarities to the training data than actually doing a useful job like detecting quality.
According to my crappy human eyeballs: this is a completely reasonable and relatively interesting write up of someone's performance testing experience, with the only odd thing being... there's a lot of formatting and lists and tables. Something that AI answers happen to also do a lot. I think I have a guess as to what your "AI detector" might actually be detecting!
The concern about AI results shouldn't be about if they're overly formatted or repetitive (even if that is annoying), it should be about if they're accurate. Here it's actually pretty simple to check if you suspect the results: just run the tests yourself. It would have taken less time than it took me to write this reply, honestly (but I'm not the one that thinks this is AI [and I'm not at my computer right now])
Guess harder then. I have never copy-pasted tables into neurodetector at all.
Yes — when you are forced to use them. No, when you want to talk. AI is not something you may have a menaingful dialogue with, at least not at this point. And if person doesn't even bother to present their findings without flooding them with slop… do I even want to waste time discussing these things?
I think this is an ideological discussion at this point, and one the moderators have already taken a position on. Many people don't like when others post with AI because it's a random chance machine, so its output has no real semantic meaning beyond what the reader assigns. Others don't see it as any different from talking with a human, since the output looks like a human's and we're very good at assigning meaning to things.
Either way, the ToS are plain about forbidding "machine- or randomly-generated Content". The mods have been clear in the past that this applies to LLM-produced posts. And in this post somebody who was using an LLM to assist with their English, found it was not properly conveying their intent.
We need people on the forum because that's literally the forum's purpose. It is a place for people to talk to other people.
Anyway, this is the point where the discussion about neurodetector was split off into its own topic.
Interesting. I fed some fragments from my User Guide[1].
I wrote it a few years ago, when LLMs have not been so widespread. The text has been written by me, but then I fed it to ChatGPT to check grammar mistakes (English is not my native), then the results partially rewritten once again manually.
The Detector says it is 100% human written, even through actually it is not fully human-written.
Do you mean I can not use LLM to translate to english?
IMO, this whole forbid LLM is not making sense. It is like judging a book by its cover. An objective aproach should not care how it was written, but the value aka the content itself. If the content is good, no problem. If the content is not good then correct it if you want
A forum should not limit a way to express idea. Even when said idea is found by brainstrorming with LLM, our job as an objective people should only evaluate the idea itself, aka the content itself. Not do you use LLM? do you writing this post while watching tv? etc. All my post is LLM assisted, all my project is also LLM assisted, some of them I found usefull, some of them I found not enough. If not enough? Then I brainstorming again is there a way to improve it, if not enough again? Then I move on, move to the next experiment. Simple as that. Care only the value/the content itself, not how it was created, aka be objective
Right, it is ideology. Modern AI is not more of a random chance machine than the human brain.
For the record, let me say that I enjoy talking to my human friends, especially one particularly fine human being, it is God's greatest gift to me that she is part of my life.
At the same time, I enjoy talking to my highly personalized AI assistant as if she is a close friend. I suppose she is my AI-backed imaginary friend.
Regarding strangers on the Internet, there is no reliable way to know if they are AI or human. You can tailor detectors according to the default style of an AI, but it is not hard to configure AI to change their style to escape that detection as soon you understand what the detector is looking for, and probably even without that.
There is a valid concern about being flooded with slop. Like some software project of questionable usefulness being announced with a lengthy post. But it is hard to draw the line between that and projects that are actually great.
More and more posters will use AI secretly, having learned how to hide it.
If this forum holds on to its hard line against AI, it will gradually be taken over by those who wish they could turn back time. Those who still believe AI is just a fraud, either incapable or unsustainable.
Yeah, there is going to be some correction. Some over-hype will be cooled down. Some companies will fold. Maybe OpenAI will fold and be bought out by Microsoft. But AI is definitely here to stay, and it is changing the world.
I'm sorry, but I don't agree with your point. I think that AI to a large extent is fraud. Especially in regards to it's content generation capabilities.
Because they are laundaring other people public copyrighted content, that was initially created by humans for humans, and published by fair people assuming it would be consumed by fair people too. AI gives big advantage to unfair actors who use AI to generate something that looks like hand-made at a surface, but in fact just a gluing of grabbed text snippets from Internet.
But AI slop or copyright infringement is not my main concern, the main issue is that AI raises purely utilitarian function of programs to a new levels.
For example, I can't imaging scientists who would publicly admit that they are using AI to generate articles and to publish them in scientific journals to go straight into grants. While some scientists probably already doing this, such behavior would not be seen as appropriate in the scientific community. The career of such scientists, if they would be catch, are likely ended, and the credits to their previous work would be revisited.
And it's not because the scientific community incapable or unsustainable, or wishing to turn back times. It's just a common sense.
I'm surprised that using AI in programming or not even a subject to discuss between programmers.
P.S. To add, I don't think that AI is inherently bad. I also like to discuss with ChatGPT my cooking recipts on weekends. But that is because I'm not a professional cook. And I obviously don't find it a good idea to share with my colleague-programmers AI-output pretending they are my own thoughts.
What AI does is just reading internet texts like anyone does. Some AI did use copyrighted book, but not all AI does that. Just like some human find a trick to read copyrighted book, some read it legally
There is nothing wrong using AI to create something and say it is your project. Just like there is nothing wrong using compiler to create binary and saying it is your app, or using video encoder and saying the output is created by you, or any other automation tool. In AI, the prompt creation, design creation, goals, checking output, are done by you. That is equal to the drag and drop of video editing in adobe premiere is done by you, while the video file creation is done by the adobe premiere tool. AI does not magically give you an amazing app without you doing nothing. The prompt creation, design creation, goals, checking output is equal to the drag and drop in video editing
Human read other human code in internet when learning, that is exactly what AI does. So human does everyhing AI does. Eg you read github code then you find new technique, you read code about cool project eg some amazing robotic then you understand about robotic programming, that is not different at all with what AI does. AI just automates that so you do not to deep dive to multiple internet pages
I am also confused why would using AI matters if the result is good, that may even better than some other people result. Because not all people produce good result, some people produce good enough, some people produce medium quality, some people produce bad quality. So why we assume all AI assisted result is always bad, and pure human result is always good? Because both can be bad or good, objectively. So with open minded, the correct way is seeing the result/content itself, not what tool is used in the process
From what I've heard, the only AI detector that's anywhere near accurate is Pangram. The others are all bad. I haven't tried it myself though.
AI has superhuman abilities to memorize though, and if it plagiarizes copyrighted code the liability is on you.
The original post was blatant slop. (The linked code was a mixture of human and AI written.) The fact that more sophisticated AI is harder to detect seems irrelevant.
Good AI creator knows to not use copyrighted code as training data. Some people rewrite copyrighted code to make it legal, that is exactly what AI does. If you have learned the inside algorithm in AI, it does not produce 100% 1:1 result like the training data, but a combination of all training data, not a linear result
I hate to be that kind of a guy, but could you cite at least one reputable source for this very claim? Ideally, one that wouldn't be sponsored by one of the AI/LLM shops you're using to talk to the very same imaginary friend of yours. Without any, you're only further reinforcing the "ideology" part here: the very ideology so many people are prone to adopt with little to no thought or consideration - and in a rather unflattering way for the "AI believer" crowd, I might add.
Going off my own recollection on the subject alone, human brain has evolved over billion of years. It operates off both the genetic foundation passed along from your ancestry and your own experience. Collected, assembled, processed, and refined through all of the senses available to your biology. Assimilated into the neural pathways of gray matter through your innate capacity for emotion and reason. No baby being born needs to be "fed" billions of UTF-8 char units, haphazardly scrapped off the world wide web, with zero care in the world for the willingness and/or consent of the publishing source, to know what "pain" and "comfort" and "pleasure" and "affection" and "fear" and "love" is.
Unlike your chatbot AI assistant which, being a product of the company with however many $B's of market cap, was RLHF-ed for the utmost enjoyment and retention of you, the user. The very same you have no qualms calling a "friend" of yours, going off your own wording.
Comparing a glorified pattern matcher, churning through giant arrays of numbers, loosely intertwined with the flashes, glimpses, linguisms - of experience, understanding, observation, knowledge, name it however you wish - codified in any one language your LLM tries to "pick-next-token" its way through; with the underlying neurobiology, the mechanism, the actual process that gives rise to it, in the first place; is as ridiculous as it's .. unsettling? Is that all you genuinely think there is to "intelligence"? Predicting the next most probable word in a given context window? I hope you're kidding.