Humans evaluating AI... not the other way around.


so many cupcakes... so little time


As a private project, DreemPort has had the luxury of allowing Hive and Hive communities to shoulder the burden of AI-checking posts.

No one writes ON DreemPort. So the platforms WHERE they write are usually the ones who are maintaining their own communities. We receive that benefit gladly! 😀

Dreemers only share their post , which already has a home URL from their favorite platform or front-end, to DreemPort - in order to have real readers, read their work. (and hopefully comment, connect, and grow beyond what they could do on their own.)

And that is what we do, and do well. Well, we also have a little fun now and then... hehe. Treasure Hunt, anyone? lol Arggggh! ☠️😂

However, we knew the time would come when we would need to add AI checks to our current screening process, which focuses primarily on plagiarism/ copyright/ and sourcing issues. The problem : we had seen the effects of false positives and the damage that had done to human writers trying to prove their humanity. I myself wrote a post about how ridiculous it was that a 100% human-written post, with absolutely 0 editing help but my own - could score as 88% human. People thought I was joking - that 12% AI seemed "low" enough for me to not take it so seriously.

No. 12% AI on a 100% written post seems like a significant 12 point error. And also -IF my 100% human post was scoring that high, then what kind of scores would be given to writers who DID employ the help of Grammarly, Taia, and the like? It made me very nervous to start something new on DreemPort without having data that I KNEW - as a human being - was accurate.

So - we created our own experiment.

For any community that wants our actual data - feel free to reach out to us, and we can show you what we determined! It's yours to use as you like! But for the sake of summarizing our data in a way that's palatable... we will just give the broad strokes in this post.

We used the free zerogpt.com as our choice of AI check. We do have the option of checking further with the paid version of originality.ai - but for our needs, zerogpt.com seemed to corroborate OUR facts - with ITS interpretation.

Because, after all, THAT is what we want to know.


We don't want to know if we are "approved as human" by an AI checker.

We want to know if we give the AI checker our HUMAN content (that we 100% know to be)- that IT is able to score high on OUR test.

That might seem like a very small distinction to make, but it, in fact, is not. I do not need an AI checker to suggest to me how my 100% human written post should be "corrected" by its robot suggestions to make me sound more human. Following? lol. In my world, I am the human and it is the robot - and never the two shall be confused.

Feel free to skip this post - and go check our results for yourself, also.

You can access the tag #dreemtest, and it should pull up all posts that were in our test. All of our tests had rewards declined, all of our participants were paid 5 HBD for their time, and all of our participants asked Hive users to NOT upvote their posts. All of the posts should also be dated Feb 2, 2024 or before. All of the posts should also have a comment from me. If any post does not meet with these guidlines above, please do not consider it as part of our test.

EXPERIMENT:

The testers were directed to write a post with the theme "I had a dream about cupcakes" as the topic. As most AI checkers need at least 100 words to test, they were directed to make their post between 150-200 words. Their first, original post was to have ABSOLUTELY NO HELP/EDITING from any kind of programming - not autocorrect, not MS Word prompts, not Grammarly - nothing.

We wanted this first test to be 100% human. 0% chance of any AI manipulation.

The second iteration was put through Grammarly, with HALF of the suggestions used.

The third iteration was put through Grammarly, with ALL of the remaining suggestions used.

If - for any reason, Grammarly was not able to be used according to the above stipulations, they were allowed to try Grammarly's AI editing tool (which basically allowed AI to rewrite the post for them.)

Our test also had an additional portion where I asked our testers to go find a very old post, preferably before 2021, and run it through zerogpt.com. We were curious to see what those test results would relay to us.

RESULTS:

On the first iteration with 100% human-written posts, zerogpt.com reported that ALL TESTS (with the exception of one) were 0% AI. The one that did register AI, was 2.75% AI/ 97.25% human. I would imagine that this might be an instance of the person not realizing that they were using a big of editing help?

Most communities will not recognize anything under 50% AI as something with which to be alarmed or concerned. Even this recent conversation with support at originality.ai relayed the same information with their AI checker.

You can see that anything under 30% AI according to them, is A STRONG ORIGINAL SCORE. For that reason, 2.75% AI did not alarm us in the least. We were VERY pleased that this checker did in fact recognize that the posts were 0% AI - all human.

The second iteration will be divided into two conditions: Grammarly-type suggestions, and full AI re-write.

The Grammarly-type suggestions DID in fact register. The scores went from 0 to 2% on average - still extremely low.

The AI rewrites jumped IMMEDIATELY to 46.68% AI, 58.92% AI, and 87.14%. As we know that this WAS manipulated by AI, these were exactly the scores we would have expected.

For our purposes, DreemPort would take scores that reached a certain threshold and run them further through originality.ai (the paid checker) to get secondary results.

CONCLUSION

We are willing to add AI checks to our CURRENT human, and manual screening process in order to ensure that the work that we promote on DreemPort meets with our integrity. Though it will be added, it will be added in a way that does not place AI interpretation over our own. We will always strive to utilize our own human methods of conversation, evaluation, intuition, and interpretation as the umbrella under which the AI checks fall. We value human interaction, and the ability to learn, grow, and change more than computerized absolutes. However, make no mistake, we also remove, suspend and permanently ban dreemers from DreemPort for being unwilling to learn, grow, and change.

How will this be implemented in DreemPort?

This has been discussed recently. We always want to present things in a fun and interesting manner - but also in a way that encourages positive growth. We feel that there is a nice way to represent this in a way that both ensures other dreemers that each post that they're reading has met with our approval, and also allows individual dreemers to challenge themselves to use AI editing less - and train their brain to become better writers.

I'll reveal more of this in our upcoming update - and I'll also most likely use our polls for dreemers to weigh in on some of the suggestions!

But for now, I'd like to conclude with one more thought, and a suggestion that might be very helpful to those who want the right tools to do the work themselves to become better writers.

In an era where time is money and money is time, the temptation to have AI write pieces for your daily post is probably at an all-time-high. And... as always, the choice is yours. But one of the things that I've noted in my own writer's journey is this: YOU WILL GET BETTER the more you write, the more you edit, the more you read, and the more you connect with other writers. If you use the shortcut method, you'll never grow beyond your starting point.

If you need help in becoming a better writer, one of our dreemers RAVES about this application that has really strongly improved her writing. In the beginning, she wrote and submitted, and edited based on their suggestions for HER to rewrite. But a beautiful thing happened. The more she used this tool, the more it started changing her writing style AS she was writing. Now, she uses this tool less - because it WORKED.

She has grown BEYOND her original style into a better writer because it changed her way of thinking AS SHE WROTE.

AI can't do that for you.

But maybe this tool can! (and no - i'm not paid by this app. lol)

Click on that image above -and it can take you to the site... nope, not a referral link or anything. Just some help from us to you! Hope it works out!

Be sure to tune in to the new AI update that we will have in the next few weeks... should be fun! should help us grow! should keep us human! hehe


Thanks to @shadowspub, @penderis, @bluefinstudios, @kemmyb, @wesphilbin, @mypathtofire, @melinda010100, @samsmith1971, @silversaver888, @lizelle, @kenechukwu97, @jacoalberts, @acgalarza, @blackdaisyft and @dreemsteem for making the @dreemport voting power go farther! ❤️


so many cupcakes image... by me and DreemPort!

screenshot of convo - through originality.ai

zohowriter image - from zohowriter.com

H2
H3
H4
3 columns
2 columns
1 column
30 Comments
Ecency