I really appreciate how the content in this video addresses what I actually need to know, not just what I think I need to know. Thanks Matt keep up the excellent work!
@@mreflow you’re welcome thank you for the Content. I’m starting a new company and I’m trying to get my website done through AI and I’m struggling but you’re assisting in that.
I see your point man. Detectors are essentially chasing their own tails. There’s no consistency across results & that’s just on the unedited article straight from ChatGPT. If you go a step further & put some light edits on it, it’s game over for the detectors. Plus, because of the lack of consistency in results it’s unlikely someone could label (with undeniable proof) that your content is ai generated. Great vid, thanks!
Great video. One thing I noticed with your GPT-1 produced text is that one of the words it used for replacement was "o'er" for "over" which would sound totally weird unless it was a pirate writing a text about the history of telephones. Also a lot of that text doesn't make any sense with random replacements that are just not how English is written or spoken. So far it seems that Quillbot is the best tool for paraphrasing. I've been experimenting with first generating texts with chatGPT and then editing them quickly with Quillbot.
I agree. It generated pretty garbage content. But also… I don’t think you need to worry about fooling the detectors in the first place. I was showing that it’s possible but not necessary.
I'm glad I don't have to worry about being detected. I enjoy writing and I usually use chatgpt for emails or to get over writer's block. Usually I'm pretty good at quickly editing what chatgpt writes to sound more human.
It's worse. I have to worry about it as an organic writer. These detectors are full of false positives and they're only going to hurt people in the long run
I honestly like the idea that kids growing up just write essays like this, it's bs busy work! Why not just get rid of essays and have students talk in class and teach them how to make arugments in person on the spot. Creativity is a good skill to learn but essays don't teach it. Speaking to others will.
Because they want you to learn the approved "facts" rather than becoming a critically thinking individual that can make their own arguments and come to their own conclusions.
Thanx matt... useful as alwayes .. the good thing you get the info from the source where most ppl didn't look at to get sure if ai generated text bad or good .. ppl must realise that ai is a helping tool not an ATM machine where in a press of a button you get money .. most YTrs unfortunately describes the use of ai as that .. thanx alot
That’s what I did in my first video on the topic. This is the second video I made about this topic and in the first video it give false positives on my human generated content.
@Matt Wolfe oh, I totally agree. Love your explanation and your videos are on point. I nerd out on this stuff and you really help me stay on top of things. Much appreciated
@Yahya Guzman looks like some posted it in the comments. Just asking it to write with 'perplexity and burstiness'. Oddly enough, this gets 100% Human in all my trials
Great Video, Myself I Copy Paste Article to Word Review and then use the Similarity section. Generally I get a zero result. So it is new content which is great for Google.
I'm working on a video about that but I put almost as much time into my thumbnails as I do the videos themselves so it'll be quite an in-depth video. The short version is that I use a combination of Stable Diffusion with a model that has my likeness trained into it, MidJourney to get a pop of color, and Canva to bring it all together.
According to the first run through the GPT-1 correction, Bell “grew up in a crime syndicate of speech and hearing experts”. Crime syndicate? Maybe the quicker and more reliable way to do this would be for a real human to read the so-called improvement before hitting the send key.
Great video again, and very interesting! Someone else already mentioned it, but the GPT-1 paraphrasing is complete trash. It's stuff that should never be posted online anyhow since so much of the replaced words make absolutely no sense in the context. Quilbot has definitely been interesting to use and, so far, has paraphrased amazingly well. A good 98% of the stuff I have had it paraphrase for testing has been completely readable and fully comprehensible as well. Having said that... I have tested the same AI Detectors you did in this video, along with around a half dozen others you didn't mention, and every single one of them flagged various short informational 'articles' I wrote on the spot from memory and my own personal knowledge as being 80% or higher AI generated. I literally JUST typed them up specifically for testing purposes!! You are absolutely right when you say they are very very unreliable right now. lol!
And that's why the whole idea of AI detection is so dangerous. There are so many real organic writers who get hit by false positives and as the AI gets better, better writing is going to wind up triggering those false positives.
The funny part, Google makes money with AI. Their costumer support is AI mostly, their search is AI. The ads are managed by AI. Humans are there just to watch the AI. BUT we as consumers or creators are not allowed to use AI
You can ask ChatGPT it self to re-write to fool an AI detector, but ask it to keep it accurate. I tested, it worked well. If you don't ask for accuracy, it will do some crazy substitutions though.
Good video. I would've liked to see some commands inputed straight into chatgpt instead for a result right out of the oven rather than being dependent on a whole separate tool.
I could have shown that too. But I was trying to make the point that fooling them isn’t even necessary. Not give as many ways as possible to actually fool them.
As a writer, I tweak, verify, edit, and supplement the content I get from these generators. I also mention/ credit that I used AI to assist and as a resource.
@AllSeeing Activity Do you think it's cheating when you go to Google to get information that you might write about? I don't. Now, if you copy and paste that's a different story. That's called plagiarism if you don't credit the source. I use Google and the chatbots as resources. I don't just copy and paste, I tweak, add, and clarify. As I'm sure you know, a writer should credit and cite all resources. That includes AI generators. That's what I said I do.
Hey Matt, Love your videos. There was a video you post a few week ago about fooling the Ai detectors and you had the prompt in the description. It was the way to teach the Ai to write like a human. Is there anyway you can share that with me. For some reason I am unable to find that video. Thanks in Advance!!!
I put 5 articles which were written by me 3 years ago through 4 detectors. All 5 articles were found to be written by AI. How interesting when we didn't have AI writers 3 years ago.
"It must be acknowledged that the content written by AI is very useful and well-written, containing knowledge and experience beyond the capability of a human being, even an expert in a particular field. The content is clear and often presented in bullet points. These values truly benefit humans, as AI delves into the roots of a problem and presents the nuances of an issue that a person cannot comprehensively address. There is no reason to ban AI-generated content due to the contribution it makes."
Have you actually read the GPT-1 piece? It's basically gibberish. The words highlighted in pink are for the most part inappropriate. It reads like something that's gone through an auto translator and back again.
because penalized ai content that contains good information or reliable information, just a ridiculous. google won't do that, because in the near future, all digital contents will heavily depended on ai.
i think @4:07 that was not what he was trying to say and the fact you got so excited was so funny because its like ignore him and follow Matt Ewolf. Anyday Matt, Anyday.
How about marking AI text by OpenAi using some code? Put "OpenAI’s attempts to watermark AI text hit limits" on Google. I will try to put a link in the reply. Not sure if URL's work on RUclips.
My guess is it's also students who want to get around writing papers and are afraid of teachers like me who are aware of AI will check any papers that seem suspiciously coherent compared to what a student can produce in class.
I'm more interested in false positives. Has anyone tried that? I'm curious because I think that is more dangerous given that I've seen some news about professors welcoming this ai detection technology without even testing it
My first video on the topic covered this. This is my second video about these detectors. In my first one, I entered human-generated content that I wrote and got false positives saying my content WAS Ai.
While I agree, we made the same assertions about the Internet when it became a mainstream in tool. What we need to focus on is teaching digital literacy not just to the students but to their teachers, too. There's just something amusing about educators wanting to use an AI to detect AI content. Back in my day we copied from encyclopedias. 😂
@@MelindaAKraft and before that, we had Cliff Notes. Technology needs to be adopted by educators, not looked at negatively. AI is a great tool - educators should figure out how best to use it.
Nah spark notes was the fire when we were in high school. I never needed to read the book. I skimmed the spark notes, and only used the book for reference.
There is absolutely no way for a Highscool student to misuse ChatGPT for their education. That the student can reliably produce the correct answer is the whole point. If the teacher is asking the right questions the student will have to show some level of understanding of the material to even give GPT the right prompt much less detect potentially ridiculous or egregious errors.
I thought this video was teaching me how to get around GPT to make it teach me how to make a grenade at home with the tools provided, that wasn't what I wanted.
Ok. So if someone wrote and article with GPT and turned it in, does that show knowledge on a subject matter? A professor could easily debunk if someone wrote an article by simply asking the student questions about the subject matter. If teachers have doubts, they can be their own GPT-detectors…
isn't it funny that people who actually do have issues against AI and want to check if something was written by AI, need to actually use AI themselves??
Editing the output will throw off the detector, so if you are only using it to *help* you write your papers and then edit the output yourself, there should be no problem. A quality paper will *always* require human input, even if it just means clever prompting and proofreading.
Additionally, no AI detector evasion tools can help you if your teacher asks you questions about parts of your paper. The only way you would understand your paper is by at least proofreading it, if not contributing your own thoughts into it.
I did that in my previous video on this channel about detectors. This is my 2nd video on the topic. The first video gave me a bunch of false positives.
So if you feel so wonderful and powerful using Chat GPT, why not own it? I mean - it's all the rage and stuff. But you want to hide the fact that you did very little to make the words line up in some set of order. What are you afraid of? Being called out as a lazy hack? C'mon man... nobody is gonna do that. Well...
Why do they care that ai makes tools for people and not seo when themselves and every human is trained to make content for seo ranking and not humans? the human aspect is just to make it readable but the seo which google has courses and certifications on how to do is for ranking lol. its a real double standard happening with ai if anyone else noticed in a few areas. its funny. as if say the humans can do it because we wont do it as good but ai can't and vice versa. not fair lol. thats even funnier to think not fair because when that does become fair it'll probably really be not fair for us lol
Lmao I just got busted for using AI for a history presentation of about 20'000 caracters ^^ I changed some words myself and used AI tools to reword it also but apparently the AI detection tools have gotten much better since this video came out cuz the teacher told me today and I'll get the worst grade possible hehe fuck it.
Yes, agree! We also use Chat GPT differently. It can give you ideas, and you can add up your insights! Such a helpful tool!
I really appreciate how the content in this video addresses what I actually need to know, not just what I think I need to know. Thanks Matt keep up the excellent work!
Good on you Matt! Great help for the community.
Thanks!
Wow! Thank you!
@@mreflow you’re welcome thank you for the Content. I’m starting a new company and I’m trying to get my website done through AI and I’m struggling but you’re assisting in that.
@@Gr8fdead wholesome.
I see your point man. Detectors are essentially chasing their own tails. There’s no consistency across results & that’s just on the unedited article straight from ChatGPT. If you go a step further & put some light edits on it, it’s game over for the detectors. Plus, because of the lack of consistency in results it’s unlikely someone could label (with undeniable proof) that your content is ai generated. Great vid, thanks!
You can tell gpt to write it in a particular way and style that isn’t as AI like
Great video. One thing I noticed with your GPT-1 produced text is that one of the words it used for replacement was "o'er" for "over" which would sound totally weird unless it was a pirate writing a text about the history of telephones. Also a lot of that text doesn't make any sense with random replacements that are just not how English is written or spoken.
So far it seems that Quillbot is the best tool for paraphrasing. I've been experimenting with first generating texts with chatGPT and then editing them quickly with Quillbot.
It just introduces errors into the text . Must of it don't make sense so such a content will be flagged by Google. Not different from using spinners.
I agree. It generated pretty garbage content. But also… I don’t think you need to worry about fooling the detectors in the first place. I was showing that it’s possible but not necessary.
Man, i watched your vids for the last days, I also watched many others. Honestly, you've made the best channel about AI.
Thanks for all your work !
Outstanding Overview and Commentary.
MAT, that was just pure geek entertainment 🤣 😂 I am so happy I found you and subscribed ❤️
Thank you, Matt! I really enjoy your RUclips Channel. Thank you for sharing your knowledge with the world.
I'm glad I don't have to worry about being detected. I enjoy writing and I usually use chatgpt for emails or to get over writer's block. Usually I'm pretty good at quickly editing what chatgpt writes to sound more human.
Is this comment ai generated
@@CG64Mushro0m No
It's worse. I have to worry about it as an organic writer.
These detectors are full of false positives and they're only going to hurt people in the long run
good for you
@@priestesslucy I guess I'm sort of worried about false positives. People already thought my comment was AI generated, lol.
I honestly like the idea that kids growing up just write essays like this, it's bs busy work! Why not just get rid of essays and have students talk in class and teach them how to make arugments in person on the spot. Creativity is a good skill to learn but essays don't teach it. Speaking to others will.
Because they want you to learn the approved "facts" rather than becoming a critically thinking individual that can make their own arguments and come to their own conclusions.
Don't you think that after failing Bard, Google will try to eliminate AI tools to keep in the search game?
This is really good research and good advice. You have been really helpful. Keep up the good work!
Thanx matt... useful as alwayes .. the good thing you get the info from the source where most ppl didn't look at to get sure if ai generated text bad or good .. ppl must realise that ai is a helping tool not an ATM machine where in a press of a button you get money .. most YTrs unfortunately describes the use of ai as that .. thanx alot
Most do, because they cant build with it, those who can will be making that kind of money.
Thanks Matt for clarifying the AI dilemma
Another way to test those tools would be to feed human text and see if it detects false positives. You never tested for false positives.
That’s what I did in my first video on the topic. This is the second video I made about this topic and in the first video it give false positives on my human generated content.
@@mreflow thanks, another interesting video. I'm glad I found this channel :)
There's a prompt that can do this without having to go to all the different tools. Makes it a bit more efficient.
But why bother in the first place?
what is the prompt?
@Matt Wolfe oh, I totally agree. Love your explanation and your videos are on point. I nerd out on this stuff and you really help me stay on top of things. Much appreciated
@Yahya Guzman looks like some posted it in the comments. Just asking it to write with 'perplexity and burstiness'. Oddly enough, this gets 100% Human in all my trials
@@AIwithRalph awesome! thank you so much :D
I heard if you ask it to increase perplexity and burstiness that it will break out of it's typical robotic output.
Tried. Not working
Great Video, Myself I Copy Paste Article to Word Review and then use the Similarity section. Generally I get a zero result. So it is new content which is great for Google.
How do you make the thumbnails for your videos?
I'm working on a video about that but I put almost as much time into my thumbnails as I do the videos themselves so it'll be quite an in-depth video. The short version is that I use a combination of Stable Diffusion with a model that has my likeness trained into it, MidJourney to get a pop of color, and Canva to bring it all together.
@@mreflow please bring tutorial on this
According to the first run through the GPT-1 correction, Bell “grew up in a crime syndicate of speech and hearing experts”. Crime syndicate? Maybe the quicker and more reliable way to do this would be for a real human to read the so-called improvement before hitting the send key.
Great video again, and very interesting! Someone else already mentioned it, but the GPT-1 paraphrasing is complete trash. It's stuff that should never be posted online anyhow since so much of the replaced words make absolutely no sense in the context. Quilbot has definitely been interesting to use and, so far, has paraphrased amazingly well. A good 98% of the stuff I have had it paraphrase for testing has been completely readable and fully comprehensible as well.
Having said that... I have tested the same AI Detectors you did in this video, along with around a half dozen others you didn't mention, and every single one of them flagged various short informational 'articles' I wrote on the spot from memory and my own personal knowledge as being 80% or higher AI generated. I literally JUST typed them up specifically for testing purposes!! You are absolutely right when you say they are very very unreliable right now. lol!
And that's why the whole idea of AI detection is so dangerous.
There are so many real organic writers who get hit by false positives and as the AI gets better, better writing is going to wind up triggering those false positives.
The funny part, Google makes money with AI. Their costumer support is AI mostly, their search is AI. The ads are managed by AI. Humans are there just to watch the AI. BUT we as consumers or creators are not allowed to use AI
At last someone talking sense on the subject.
well people were creating auto generated content from a long time only now it is more great and helpful!
Chat GPT is just a word calculator. Write an essay in class to get tested.
Great vid, good value info.
You can ask ChatGPT it self to re-write to fool an AI detector, but ask it to keep it accurate. I tested, it worked well. If you don't ask for accuracy, it will do some crazy substitutions though.
Amazing info and resources.
Good video. I would've liked to see some commands inputed straight into chatgpt instead for a result right out of the oven rather than being dependent on a whole separate tool.
I could have shown that too. But I was trying to make the point that fooling them isn’t even necessary. Not give as many ways as possible to actually fool them.
As a writer, I tweak, verify, edit, and supplement the content I get from these generators. I also mention/ credit that I used AI to assist and as a resource.
@AllSeeing Activity Do you think it's cheating when you go to Google to get information that you might write about? I don't.
Now, if you copy and paste that's a different story.
That's called plagiarism if you don't credit the source.
I use Google and the chatbots as resources. I don't just copy and paste, I tweak, add, and clarify.
As I'm sure you know, a writer should credit and cite all resources. That includes AI generators. That's what I said I do.
Ok, what about plagiarism? For example, writing an ebook might give a plagiarism problems
Hey Matt,
Love your videos. There was a video you post a few week ago about fooling the Ai detectors and you had the prompt in the description. It was the way to teach the Ai to write like a human. Is there anyway you can share that with me. For some reason I am unable to find that video.
Thanks in Advance!!!
I put 5 articles which were written by me 3 years ago through 4 detectors. All 5 articles were found to be written by AI. How interesting when we didn't have AI writers 3 years ago.
10:05 did you actually read the content it made? completely bricked it.
he grew up in a crime syndicate of speech and hearing experts 😆🤣
I did. It’s garbage. The point of the video is that it’s not even necessary.
Astonishing content Matt 💯💯💯 Thanks to share your knowledge with us. Thumbs Up from Brazil 👍👍👍
Even just running AI text through Grammarly and then sense checking it can make a difference.
12:26 - I think its more that people use this for school and don't want their teacher to run it through an ai detector and get in trouble
Thanks for this Matt
amazing, thanks for sharing! 🔥
Ask chat gpt to write content that looks humanly and not robotic and it would do that
"It must be acknowledged that the content written by AI is very useful and well-written, containing knowledge and experience beyond the capability of a human being, even an expert in a particular field. The content is clear and often presented in bullet points. These values truly benefit humans, as AI delves into the roots of a problem and presents the nuances of an issue that a person cannot comprehensively address. There is no reason to ban AI-generated content due to the contribution it makes."
Great video about this!
Have you actually read the GPT-1 piece? It's basically gibberish. The words highlighted in pink are for the most part inappropriate. It reads like something that's gone through an auto translator and back again.
I have. It’s garbage. I don’t think people should try to fool the detectors in the first place. I made that pretty clear.
I think OpenAI's detector is less reliable simply because they prefer to err on the side of caution.
Great info!
I was messing with the detectors and all you have to do is add in a couple of typos and boom it’s “100% human written” lol
Great summary. You're right the content just has to be helpful.
If you copy the paraphrased content and repaste it on chatGPT with a one liner: "Was this written by you"? ChatGPT catches it.
because penalized ai content that contains good information or reliable information, just a ridiculous. google won't do that, because in the near future, all digital contents will heavily depended on ai.
i think @4:07 that was not what he was trying to say and the fact you got so excited was so funny because its like ignore him and follow Matt Ewolf. Anyday Matt, Anyday.
Images, art? Can you help us get around detectors with images if there are any?
How about marking AI text by OpenAi using some code? Put "OpenAI’s attempts to watermark AI text hit limits" on Google. I will try to put a link in the reply. Not sure if URL's work on RUclips.
I fucking love you, i have been looking for this solution for tons of hours, thanks 🇲🇽
My guess is it's also students who want to get around writing papers and are afraid of teachers like me who are aware of AI will check any papers that seem suspiciously coherent compared to what a student can produce in class.
how to bypass chat gpt limitations
I'm more interested in false positives. Has anyone tried that? I'm curious because I think that is more dangerous given that I've seen some news about professors welcoming this ai detection technology without even testing it
My first video on the topic covered this. This is my second video about these detectors. In my first one, I entered human-generated content that I wrote and got false positives saying my content WAS Ai.
Are there detectors for AI Art? Is this relevant and if yes, what are the ways to fool those?
Nice share ❤❤
Can you tell ChatGPT to use more "burst" and have it show more human?
Yea. There are an almost infinite way to fake out the detectors but why you’d need/want to is the real discussion, in my opinion.
You are under estimating the number of highschool students that misuse this
schools are useless anyways
While I agree, we made the same assertions about the Internet when it became a mainstream in tool. What we need to focus on is teaching digital literacy not just to the students but to their teachers, too. There's just something amusing about educators wanting to use an AI to detect AI content. Back in my day we copied from encyclopedias. 😂
@@MelindaAKraft and before that, we had Cliff Notes. Technology needs to be adopted by educators, not looked at negatively. AI is a great tool - educators should figure out how best to use it.
Nah spark notes was the fire when we were in high school. I never needed to read the book. I skimmed the spark notes, and only used the book for reference.
There is absolutely no way for a Highscool student to misuse ChatGPT for their education. That the student can reliably produce the correct answer is the whole point. If the teacher is asking the right questions the student will have to show some level of understanding of the material to even give GPT the right prompt much less detect potentially ridiculous or egregious errors.
Hi Matt, I submitted my email, 2 weeks ago, for your Friday newsletter (twice) but no response. Any suggestions. Excellent RUclips channel!
WordAI is capable of rewriting any article that passes the AI detectors.
How can you claim the detectors do not work when AI cheat check caught the AI text each time until the text was edited?
Awesome 👍👍
🎶thank you🤗
that was a very long steps you took, just to extend the length of video!
Gpt minus one is great
It’s seems as tho gpt-1 adds in errors to make it appear legit so not sure how helpful that is 🤔
Depends on your purpose for trying to fool them. If you’re doing it for SEO, fooling them is a waste of time anyway.
I thought this video was teaching me how to get around GPT to make it teach me how to make a grenade at home with the tools provided, that wasn't what I wanted.
i see you asking "why?"
Here is the answer:
College
Ok. So if someone wrote and article with GPT and turned it in, does that show knowledge on a subject matter? A professor could easily debunk if someone wrote an article by simply asking the student questions about the subject matter. If teachers have doubts, they can be their own GPT-detectors…
@@mreflow yeah but when they ask you to write 29384 pages about something you basically know, then chat-gpt becomes so handy.
isn't it funny that people who actually do have issues against AI and want to check if something was written by AI, need to actually use AI themselves??
Thank you for this ^_^
Just change the letter L when in lower case to a capital i. twice. all good.
anyway I just watch this vid cause want know if my teacher could know I using ai help for finish my homework by use gpt detectors :)
Editing the output will throw off the detector, so if you are only using it to *help* you write your papers and then edit the output yourself, there should be no problem. A quality paper will *always* require human input, even if it just means clever prompting and proofreading.
Additionally, no AI detector evasion tools can help you if your teacher asks you questions about parts of your paper. The only way you would understand your paper is by at least proofreading it, if not contributing your own thoughts into it.
Quillbot is pretty good. Gpt-1 is trash, the synonyms it uses are ridiculous and don't make sense.
gptzero is pretty hard to fool though
You sould test with real text if thoses detectors find them generated by ai.
I did that in my previous video on this channel about detectors. This is my 2nd video on the topic. The first video gave me a bunch of false positives.
Thanks \m/
what about ROMANCE words?
Just as ChatGPT et al is very likely to improve over time, the same can be true for AI Detectors...
Yep. Exactly what I said at the end of my video. But why do we need to fool them?
Then Chat GPT will become so good where it will be impossible
how about Gptzero?
GREAT!
You are so good😊
I see a lot of students using this.
SIR find me ai, where I can put a pic and make a 3d character / 3d building / 3d item
Sure, GPT -1 can fix your detection issues but did you read that article? It basically became nonsense lol
So if you feel so wonderful and powerful using Chat GPT, why not own it?
I mean - it's all the rage and stuff.
But you want to hide the fact that you did very little to make the words line up in some set of order.
What are you afraid of?
Being called out as a lazy hack?
C'mon man... nobody is gonna do that.
Well...
Why do they care that ai makes tools for people and not seo when themselves and every human is trained to make content for seo ranking and not humans? the human aspect is just to make it readable but the seo which google has courses and certifications on how to do is for ranking lol. its a real double standard happening with ai if anyone else noticed in a few areas. its funny. as if say the humans can do it because we wont do it as good but ai can't and vice versa. not fair lol. thats even funnier to think not fair because when that does become fair it'll probably really be not fair for us lol
Lmao I just got busted for using AI for a history presentation of about 20'000 caracters ^^ I changed some words myself and used AI tools to reword it also but apparently the AI detection tools have gotten much better since this video came out cuz the teacher told me today and I'll get the worst grade possible hehe fuck it.
DOPE!
QuillBot works? What the fuck is this, with me it doesn't work.
Not good, that menas google can deploy lamda tomorrow and its over for everyone, so keep it up
FOR GOD'S SAKE PLEASE USE THE DARK MODE!!!!!!
NO! Everybody can't be like you.
Wait!! Is Matt AI generated????
trying to string a copy paste prompt out for 15 minutes is what gives technologies a bad name. Stop the useless stories and get to it
ZeroGpt it's good
Try copyscape
hehe 😃