For those wondering about a text that was written manually. I put my PhD thesis literature review into originality and got a 1% AI score. Make of that what you will!
Well done for your exploits of AI as an academic writing tool. I'm still waiting for that book you haven't written yet. i.e., "Completing your thesis in 6 months using AI tools". I bet once the first school summons the courage to adopt AI as an ordinary writing tool, your book will be handy for the ambitions. These tools are long overdue for research purposes and in aiding the physically challenged to contribute to the body of knowledge (imagine what text-to-speech could have done to the likes of Professor Stephen Hawking, of blessed memory). At the end of the day, it is original ideas that will triumph.
Andy, I also wanted to thank you for you hard work. You have inspired me on many mornings when I just did not want to write. I watch a video and I want to work.
I actually scanned my proposal paragraphs -which were written before GPT existed - and it detected 98% AI... so I wouldn't say it's very accurate unfortunately.
I use AI for framework and direction/guidance in bullet points and then research each bullet point to write in my own language. This works. AI is your assistant to write not write on your behalf.
I am wondering why this guy is typing in the percentages. I pasted a text that was AI-generated in this "detecter" and it came back with human. I did this more than once. I would not trust this. But if you use A I as a guide and then research yourself, it would be good.
I condense about half of the content in my own words, which often leads to favorable grading from my professor. This approach saves time on reading and researching. While I recognize that it's primarily the professor's responsibility, the industry heavily relies on AI for various tasks such as policy writing, coding, and drafting letters. Those who fail to adapt risk falling behind. As a side note, this message has been enhanced with AI.
I think it's actually fine to use AI to produce academic work. AI doesn't generate original ideas, it just implements them. Think of chat-GPT as a lab assistant or something. I have used it to generate the code needed to run my regressions nevertheless. Since I don't really like the way AI writes academic work I don't use it that much in the paper, but since in my field all experiments are statistical I use it heavily as an intern 😂
for the last 40 years (at least) the academic world "relied" on enslaved research assistants to write articles that were signed by the dept director (and sometimes the research assistant was not even mentioned in the authors list), and this has never been a scandal... But now, all of a sudden, only because it is AI writing it, it is becoming a big problem. Where is the problem/scandal if you generate your data, you generate your analysis, you come to your own conclusions and after you provide your guidance to get chatGPT to contribute to or assist with the writing of the final paper...? I understand that we should define the boundaries of "contribute to" and "assist with" but for me, introducing AI in the flow of actions in research could be more an opportunity than a disaster. I agree with you --I don't see that having an AI copywriting support is an issue (as long as you, not the AI, went through the phases of the research)
Keep in mind it’s difficult to define what originality means, and that originality and creativity are emergent properties of intelligence. More powerful models will surely be capable of both, and I suspect that had you seen a glimpse into 2023 from the past you might have thought the results of LLMs were creative and original.
I have very mixed experiences with these detectors. I tried to put my text into HiveModeration, ZeroGPT, GPTzero, Writefull, Copyleaks, AI DETECTOR, AI Text Classifier, and now OriginalityAI. I tried to put two versions of the same text into them. The first version is raw text in latex format (with citations and equations in Latex format) and the second version is text copied from pdf. Some of them show that there is some AI written text, some that it is original work, and some of them with one version showing it is original work, but the other version is not. OriginalityAI shows it is AI written text most of the time based on one word or phrase, which is general. Most of them label the text below the equation explaining each variable as AI-written or as plagiarism. The most crucial thing is that they don't understand how citations work. They show sentence is written by AI because they have maybe seen it during their training and ignored that there is a citation right after that sentence. Showing that the original text is written by AI when it is not can be very dangerous.
Undetectable is rubbish. They seem to think that by inserting lots of punctuation and grammar errors, they will be more human. One of its favorite tricks seems to be something like this: 'The field of "Compliance Management" plays a role, in guaranteeing that activities comply with laws, rules and industry norms.' Note that comma after 'role'. I found that it created this structure in almost every input that I gave it. Did you actually read the output? How could you recommend something like this?
This is not true AT ALL. I put in my papers from my undergrad from 2014 in this Undetectable Ai and it told me that it was written by Ai 99% which is just rubbish since Ai was not around at that time.
Same thoughs actually, because those things might not work well because every single human is unique and they're style of writing too! And you can't just blame a person for using AI just because his text looks robotic 🥴😒
Originality gained soooo much input data during the past few months that it is now saturated... so anything is scanned as AI. I believe it's done. There are others tools to detect your texts. Try to use more than one, compare and adjust.
AI has gotten to the realism where AI detectors can tell when AI looks like human writing, but can't tell when human writing is actual human writing. If that was too wordy for you, AI detectors will tell you human written articles are AI even though they aren't. That's how good AI is.
If you want to remove AI detection you have to read the text very carefully and remove all the words that doesn't make sense. Those AIs are not very good at writing. They use nonsense words in the texts. You may not notice it during the first reading but if you reads the text very slowly and carefully you will recognize those words. Remove them and replace them with synonyms. Also pay attention to the grammar especially at the end of the sentences. Those AIs use weird grammar and end the sentences at a strange way. It doesn't get detected by grammar checkers. That's what I do to remove AI detection. Sometimes you can't get rid of the AI detection especially if you use bullet points and wrote parts of the sentences. They get detected no matter how you write them.
I can't help but think we are going about this wrong. Why not use the detection of AI to augment understanding rather than flunk people. And, as an AI researcher, I can tell you that AI detection is about as good as a coinflip. Especially with the new prompting techniques that students and others are learning. Rather than a cat and mouse game, why not use AI detection as a signal that students may need additional help? AI isn't going anywhere...
Not very scientific. Not constants, only variables on one sample. Another question is maybe the ai detector isn’t good. How can we trust that? Fun experiment though, just not all that conclusive, broadly speaking.
If paraphrased work is being detected as Ai then it means that AI detection tools are way too harsh now and they are punishing slightly original content as well
Not only slightly original but completely original - and that is the purpose of AI, viz. to create sufficient uncertainty in us and to arrogate all thought and power to AI. Paranoid? You bet!
I would absolutely challenge any AI detection to my work. If detection tools are giving out "97% AI" for paraphrased work, then they're clearly not fit-for-purpose. I dont think any student or academic will be failed or rejected on the strength of AI detection tools alone. Plagiarism can be proved without any shadow of a doubt. It is easy to build a really strong, irrefutable case for plagiarism...the same cannot be said for AI generated content. There will always be room for doubt. Someone may just have a writing style that is very similar to a generative AI tool. There is no way to definitely prove that a piece of content has been generated by AI. Institutions would end up being sued if they start heavily penalizing people on the strength of detection tools.
People with Asperger’s Syndrome and Autism write very similar to AI. Since AI came out, I have been accused of being AI. I am not. Just Asperger’s Syndrome.
Especially in marketing, because AI has been loaded with a lot of marketing and blog writing styles. I have an about page article that was written for one of my websites that came back with parts of it. AI generated up yo 57.2%. So now I'm trying to figure out if it's just the writing style or if the writer used chat gpt. The detector I used is copyleaks, the best one I've found so far
I'm new to the AI thing but I write books. I think you are right, the AI tools bomb everything out and perhaps if you pay for them, maybe they will give you a pass. I don't know but even with rewriting the AI stuff it detects it as 100% AI. Something is fishy here I think.
it's sad......what if someone is a good writer? Good with grammar? And so on......these AI systems are giving false positives to convince people to buy their "premium version" or more "tokens".......make it make sense....... there is a war on AI software programs that don't have people's best interests at heart.....its all a game
I'm a freelancer but I have to stop writing essays coz of these AI detection tools. I hope guys making these algorithms are aware that wrong predictions can really end careers or even cause bad grades and they are at fault. Do i have to write dumb to pass an AI check?
Ai detection doesn't work. Everytime you get a 100% score, question it. I copy-paste old articles from 2008 into ai detection and it was 100% AI. So yeah, AI detection is just not working. Try it yourself.
Originality is total garbage. I purchased a subscription there and the way it works it's just wrong. First, I put 100%AI content and it was found. Then, I put mixed content and it flagged it as 100%AI. Then, I put content from my master's thesis from 2017, 99% AI. Then I wrote a paragraph 69% AI. I edited the grammar and spelling mistakes with Quilbot, 100% AI. This shit doesn't work. It just flags almost anything as AI despite being written before even AI was a thing.
I put a text that I wrote in 1999-2000 and gave me 40% ish AI, then I submitted the same text to chaGPT for grammatical and syntax corrections ONLY (as I am not an English native) and resubmitted it to Originality and WinstonAI it gave me 99% AI. The only ones that gave me a more accurate classification of the text are openAI Text Classifier and Unicheck. None of the other was able to make a difference between text written by AI and text corrected by AI
I actually tested papers from 2015 in several AI detection bots and they were flagged as AI... I am in college so this has become very stressful for me to have my essays flagged when I know I wrote them without the help of AI.
AI detectors like this are not accurate, different company produce different scores. The results of my test, all of my Google translate results were detected as AI even though I wrote the original text from scratch. Even when I wrote several paragraphs of nonsensical fictional stories. I tried to translate my thesis to English using Google translate, it was detected >90% AI, even though my thesis was made from scratch and was a super niche topic. My thesis was written in 2010, when AI didn't even exist. 🤣
I have an 100 point essay I have to turn in and I'm scared I'll get failed for it (my teacher has already failed 2 students for it due to the ai checker “turnitin”) bc I worked really hard on it, but I've put it through ai checkers and all come back saying it's 40+% ai. ☹️
I typed in "how are you" into an AI detector and it says 100% AI. Right there I knew these detectors were actually fucked up. Their outcomes are based on how they feel about a particular sentence or phrase. 😅
The Originality AI seems overly sensitive, skewing towards false positives. I tested one of my original articles from the early 2000s, and it labeled it as 79% AI-generated. It appears to default to marking everything as AI-produced unless proven otherwise. It's a case of 'guilty until proven innocent'
So...2 things: 1) The percentage shows how confident the software is that the tested text was generated by AI; NOT the percentage of text that was generated by Ai. 2) I just ran a paragraph of my PhD thesis through it. I 100% made those lines of text because it is (literally) my original contribution to the field of IR. The results came back 50/50...First impression: This is a dangerous tool if professors think it is the end-all be all to detecting Ai...a lot of students might lose that battle just because the professor can't see my first point and into my case study (which proved to be very not accurate)
Exactly, these tools don't work. There is no way to detect 100% AI content, and we need to hold these software vendors to firm explanations of what they're tools. There is absolutely no way to detect these thing. LLMs work but using probability and other factors based on HUMAN written language. OpenAI is a closed system, even they couldn't create a tool that can tell 100% ai written content. These vendors are going to create a big issue for academia and professionals.
This reminds me of the case where the professor flunked the entire class for using ChatGPT to write the essays, even though they did not. Seems his tool was flagging everything as "AI detected" regardless of whether or not it actually was written by AI. 100% false positive! How do we know that's not what's happening to you? Perhaps any well-organized grammatically correct text is flagged as AI.
Andy, great video, as usual ! Many thanks. Please, note however, that most of these tools are not able to make a difference between text created by AI and text corrected by AI, and will give all of them a low human score. This particularly impacts non English native people that tend to use AI tools to correct their texts. (I know it by... personal experience !) Based on my own experience, UniCheck gives a very different score to text generated by AI and text corrected by AI, which could be because it is capable of making that difference.
@@Thischangeisnottelling I ran several (non scientific) tests over the last weeks using different texts across almost 10 tools and found that openAI's AI text classifier and zeroGPT are the less ambiguous (which does not mean the more precise...). On the other end of the spectrum, UniCheck and Winston went wrong almost every time they were submitted a text corrected by AI
There is absolutely no way to 100% determine if something is written by AI. I even passed an old paper through this and it still gave about 60% confidence that it has been written by an AI. And the last time I checked, I was very human. I'm a computer science grad so I've been trying to crack what patterns these detection tools look for determining AI generated content. The level of false positives given by these tools leads me to believe that there is no quantifiable method that can be used to detect AI generated content. Accuracy of AI detection is also substantially worse than the accuracy of information in AI generated content. Detection tools should never be used as a grounds for dismissal in academia, given how they are biased to give out a false positive. Being fairly new, there hasn't been any acceptable form of large scale testing for the detection tools. Any product that hasn't been tested with large datasets by multiple third parties, should not be implemented to determine futures of students.
100000% correct. We are now using algorithms to detect other algorithms. Should we create another algorithm to detect those algorithms detecting each other? This is nonsensical, how stupid are we getting at this point?
I think the AI detection scene is a bit of a mess. I put my own text that I wrote 2 years ago in chat GPT AI detector and it told me it's 100% AI written
And you're right. These things DO NOT work. There currently is not a single known AI model or system that can produce better _quality_ in a task than an equivalently trained human. Image recognitition, writing, writing detection, etc. - more generally, perception, creativity, and reasoning. You see things like "our proposed AI system outperforms human readers by 3%" or whatever in a medical imaging study, which is impressive, but you have to contextualize it within a highly currated dataset that in no way captures all of the variability observed in the world. These models up to now always fail on adversarial examples, while humans do not. And so my point is, if humans cannot reliably detect AI-generated content (except maybe Noam Chomsky), there is no model in the world (looks nervously at Google) that can reliably do so either. And the reason ChatGPT is so disruptive, is because humans cannot reliably classify the content it generates. This video was an ad.
It appears from what is written here that, using Grammarly to correct errors in my writing, have turned it into potentially AI written. So, still going to use Grammarly just only for spelling and typos, and not their sentence suggestions.
Ohhh yes it's true if you used Grammarly or such as app saying potentially AI....you own sentences and mistakes fixed by app and saying potentially.... it's true
Chatgpt doesn’t write well enough for academic papers anyway. It’s great for providing a general structure that can be heavily edited and elaborated or for getting started just to build momentum.
and that's the best way to do it, instead of asking it to write as a whole :), but also fix reviese and edit it, ask ai to use alternate words according to your understanding. In that sense you make it your own from ai generated text.
I put my original work on Orinality AI and it detected as 100 per cent AI... It's not reliable. I used to like your content but lately, I feel like you started to use your channel only for the sake of advertisement. Your content becomes unreliable, what a waste.
I am a 35 year old man. I am a 35 year old man. One of those statements was written by an AI, and the other written by a human. Good luck differentiating which one.
All is just they are selling their product. I am writing a text with my own, and the software detects 99% Ai. All these companies are interconnected to buy one software than another software.
So I did my university thesis about 10 years ago when any of this stuff hadn't surfaced yet and for fun run it through one of these AI detectives and got that my own hard work, blood, sweat and tears (I wrote that bloody thing for 8 months) is 80% likely written by AI. Like what the actual fffffffffff 😮 am I an android? 😂
One big issue is that simply correcting 100% original work and running it through Grammarly can show up as AI. This has been happening while I have been writing comps for grad school. I am worried that schools will use Originality as some sort of definitive answer, when it seems to not link scientific writing that has perfect syntax and Grammarly corrected. I have multiple examples saved on my computer that show a before and after with Grammarly being the only difference, so that I can defend myself if I am accused. side note: I rean this for AI: it came back 94% likely human. Got this from grammerly: One big issue is that simply correcting 100% original work and running it through Grammarly can appear as AI. This has been happening while I have been writing comps for grad school. I am worried that schools will use Originality as some definitive answer when it seems not to link scientific writing that has perfect syntax and Grammarly corrected. I have multiple examples saved on my computer that show a before and after, with Grammarly being the only difference, so I can defend myself if I am accused. now its 53% likely AI. take away... Originality is a false positive machine. Just to pass its "detection" you have to write worse on purpose. They are going to lose a bunch of lawsuits if students get kicked out of programs as a result of their tool.
Just came on to say this as well. I took a sentence that I wrote. Originality identified it as original, but I used Notion AI to check it for errors. It didn’t make any changes and it flagged as 100% AI. This is a terrible tool.
@@seanaevelyn7518 it is a false positive machine. What’s funny.. you can take something as 100% AI, mess with it one word at a time, and you get one little change that will turn it to human. It makes no logical sense.
@@UniversityAIed no,. Simply using the checker that fixes syntax and gives suggestions on how to improve clarity. The tool that have been available for a long time
This is so true. I have tested my work and a few of my son's middle school essays. Original work 95% human, then with grammarly 100% AI. I have come to the conclusion that originality AI is to grammar sensitive to use for anything.
originality is very sensitive and will flag everything as AI generated I put in the federalist papers in it and it said 100% AI generated so alexander Hamilton was using GPT back in 1760s it’s pathetic
Pls follow up this video with a broad range of original texts and original-AI-mixed texts and how those checkers perform there. Thx. Great work you are doing.
In my experience so far, I check paragraphs that show 90 plus porcent of AI when I manually paraprhased it entirely. This is gonna bring trouble because often times will appear to be AI generated when it is not. We'll see.
I'm dating myself here, but I remember when I was in the 7th grade, calculators were a complete no-no. Writing essays with AI reminds me of that-just my thought.
Unfortunately, that’s going to be nightmare for students who is even writing by themselves without any using gpt but still detecting as ai ,academic will not understand that till 3 years will pass
This app is the biggest scam ever man. And this guy seems pretty confident , i just uploaded my original work on it and it says 100% AI . To those of you reading instead of going for these paid websites just use the free zero gpt detector that does a better job TBH.
If writing using AI tools is considered wrong or taboo then the same logic of thinking should also apply when using AI to detect AI-assisted writing. Based on this, using AI to detect AI is very wrong, in addition to creating many possibilities of wrong accusations and causing anxiety that should not exist.
undetectable does not work. The only reason it works is it outputs a huge block of virtually unreadable text with little punctuation. No one is going to publish that without inserting paragraphs and punctuation. and guess what...do that and it won't pass originality. So in summary, it simply doesn't work.
What I’m not getting, is he said he rewrote something in his own words from start to finish, so in short he wrote his interpretation of the AI, how is this different to normal paraphrasing, and is there a chance AI is scoring things higher than it should, like where is the evidence to support the algorithms are correct. People are putting a lot of faith in other companies algorithms which aren’t really proven. It’s getting to be like music, where there are only so many notes that work together, equally there are only so many words that make sense together, as more and more people use these tools to rework their own material, there is a risk AI will add this to the language models incorrectly. Eventually these AI tools will be no different to spell checker in WORD.
I have a question/ problem; the issue is, when you generate a text with chatgpt, then paraphrased by any tool, once you copy the paraphrased text and paste it into chatgpt again, it will detect the text. So far none of the websites/tool for paraphrasing works. have you tried this?
I put in one of my old papers from 2019.... detected Ai. Sorry. These checkers have been proven to be wrong. To the point that schools do not allow their use. Try writing a childrens book, the language you use to write mimics Ai language because of the simplicity of language.
Agree with other comments about the paraphrasing bit. Paraphrasing of done correctly is NOT plagiarism. And also should not be considered “ai”. So if you paraphrase ai, a book, a podcast, or whatever else. It IS considered legit and should not be considered the original.
So let me understand here: Using scientific calculators are ok (because of our fear of math skills) but copying some random paragraph for English language (which has zero business value according to me) is a big no no. Hypocritical.
I agree with another comment that this video is most likely an ad. Andy is tempted considering his large audience since his experiment is biased. It does not include running an entirely original text thru the detectors. Most likely such text will also be marked as produced by AI. If institutions start using those detectors, most likely we will need to start making mistakes on purpose. The hope is that academia will embrace the inevitable technological changes, accept the idea of using technology as help and start adjusting assignments and the way students’ work is evaluated. As for dissertations, I hope that the originality of the scientific work is what is actually important. At least in STEM disciplines.
@@DrAndyStapleton it kinda screams like an AD especially since u didnt mention that you added referral links to the services that you say FOR SURE ARE THE ONLY THINGS THAT WOOOORK. You get paid each time someone signs up using your link. Seems a bit suspect.
They are not this accurate. This looks more like an add for Undetectable AI. No offense intended. I've experienced false positives with almost all the checkers and have been able to prompt engineer a human rating from all of them. Also, there are some out there that are really bad and just call everything AI generated.
Hi Andy, thanks for the video. We would appreciate if you could make a video like this about text corrected or paraphrased by AI like chatGPT. I mean using ChatGPT like grammarly and Quilbot, but not to generate text.
Thanks for the insightful video, but a quick question, please. Will my content count as AI content if I create content and use Chatgpt to paraphrase it for fluency? In other words, will the paraphrase of my content by Chatgpt count as AI content? Thanks in anticipation of your scholarly responses
It seems obvious these tools are not yet good enough to use for academic plagiarism checks. Making the AI restructure your own writing output 100% does not seem to me as fitting for plagiarism checks. I mean, you literally wrote it, the AI just restructured the text to be easier to understand and read. In addition, I will assume spelling tools like Grammarly actually will do more similar things later on. As of now, It does not only help with spelling, but sentence structuring etc. If that is not similar to AI then idk.
If the product is correct and an accurate reflection of the author's views, the literature cited, and the results of the original research, then who cares if an ai, a person, or some combination thereof created the manuscript? If it is good enough to pass peer review, that's all that matters
I can get pass it 90-100% Just think its simple. Ive took 100% A.I content from chat gpt as well as A.I news feed websites. I then added a simple prompt and presto. " Just say A.I thinks of you as an inferior being!!!!!!!!!!!!!"
Something I've been doing is - I will use OneNote to dictate what I want to say, and then ask ChatGPT to fix the grammar and punctuation errors from the dictation software. Then I can usually ask it to improve the item "while keeping my personal style of writing". This usually gets 0% detection on GPTZero. To be fair though, at that point basically all the content is already mine. That is my way of getting past my "blank page" writers block.
I am Also Confused About This AI Detection Stuff. I watched your video because i TRULY believe AI writing is a scam and I write everything 100% myself. HOWEVER: on average, after testing like you did in the video, with 12 of my 100% human written articles with an average of 1200 words, My Average AI Score was 47%???? WHAAATTT??? It doesn't make sense... If my original written articles are being flagged as AI, than WHAT is REAL???? Some articles had even as high as 70-80% AI Detected content.... I wonder, are the AI Detection Platforms DRIVEN by AI?????
Great video! I feel like while people are just discovering the advantages to use AI for academic work, there is a small gap to even take advantage it. The speed of technology & AI is just too fast.
Just gone through your comment, I want to suggest to use another tool @stealthwriter. I have been using it for multiple writing purposes so give it a try I'm sure you are going to like it.
Im writing manuscript to publish in journal, can you give me advise how to paraphrase from others article sentences without get cought by sophisticated ai detectors
As a GenZ who's working in IT and has a degree in comp sci and comp eng. this is like saying I used calculator to do my work. It's either gonna get embraced now or it will be when our generation comes along in the management level.
The output is without grammar errors. Unreadable. When you use grammar corrections tools . It gets detected as ai generated. If you tried to edit the document yourself. It still gets flagged as ai content
Here is a problem as well. I have written completely original works and put it through detection and sometimes it will come back as AI generated. BTW, cursing is a good way if you are using Chat GPT as it won't generate vulgar language. As my guy here is saying use it as a tool. Beware though, run your own work through the detection options available because it may just think your original work isn't original, especially if you read a lot, you may have something that is considered derivative. I suppose ghost writers are going to have some problems as well. Cheers 🍺 and good luck!
Could you put some human written text into that AI detection to see if the AI detection algorithm works, or it just throwing false positives for everything?
My take is like showing your steps in math solution, you can give meta data, which are the stages of paper to prove its veracity. It is something like what we called footnotes.
It seems like many have had AI detected for their own writing. Has anyone managed to remain undetected for A.I for AI used writing with this tool(Originality A.I)?
I recently used an AI-powered tool to enhance my essay by correcting grammatical errors and making it sound more like a professional essay. Despite hearing rumors that relying on AI could lead to failure, I took the risk and was pleased with the results. However, my friend was caught using a similar tool and I am unsure of how he was caught.
Great video! Thanks! WOW. Very powerful. ********************** Listened to the end and I love your conclusion about using it as a tool, NOT to generate content. Very reassuring. Thank you so much!
