Hello,
What are the rules regarding AI generated reviews? I spent litterally 10 seconds to generate a perfectly written one.

Title: The Black Album: A Musical Misfire That Needs a "Master of Puppets"

Rating: ★☆☆☆☆

Prepare yourselves for a comedic disaster! Metallica's self-titled fifth studio album, lovingly called "The Black Album," is here to bring you an unforgettable experience in disappointment. This album is a real headbanger, but not in the way you'd hope. Get ready to bang your head against a wall out of frustration instead!

The Black Album, huh? More like "The Bleak Album"! Metallica took a sharp left turn from their thrash metal roots and decided to embark on a musical journey that resembles a leisurely stroll in a retirement home. If you were expecting the roaring energy and furious guitar riffs that Metallica once championed, I have bad news for you—those are as absent as my dance moves at a party.

It's almost like Metallica decided to play a cruel joke on their fans. Gone are the days of intricate guitar harmonies and mind-blowing solos that would make your fingers weep. Instead, we're treated to guitar riffs that are as basic as a "Metallica for Dummies" guide. I can't help but wonder if they accidentally swapped their guitar picks with popsicle sticks during recording.

But wait, it gets better! The songwriting on The Black Album is so repetitive that it could put a broken record to shame. I've heard more surprises in a box of plain white bread than in this album. Each track follows a formula so predictable, it's like Metallica hired a team of psychic songwriters. Slow intro? Check. Chugging power chords? Double check. Unimaginative solos? You betcha!

Let's not forget about the lyrics. Metallica must have attended a workshop on how to write the most generic and uninspired lyrics known to humankind. They explored profound topics like love, loss, and personal struggles with the same depth as a kiddie pool. It's as if they traded their poetic pens for the cheap knockoffs you find at the dollar store.

And the production quality? Well, let's just say Metallica must have confused the sound engineer for a dentist because they polished this album so much it could blind you. The rawness and grit that once defined Metallica's sound are lost in a sea of excessive production. It's like they threw their music into a washing machine set on "extra spin," leaving it devoid of any soul or personality.

In conclusion, The Black Album is a perfect example of how not to make a metal album. Metallica's attempt to broaden their appeal backfired like a firecracker in a clown's pants. If you're a die-hard fan of their earlier works, brace yourself for disappointment. You'll find yourself yearning for the days when Metallica was a master of puppets, not a jester of mediocrity.

It's a tricky question, in the sense that the site rules don't explicitly address this topic if I'm not mistaken – but I think it's bound to change, as there's been a surge of AI-written reviews sent lately and it's likely to be a bigger and bigger problem as time goes by. For now, they've all been rejected since they all were very obviously not written by humans (please check the latest posts in the 'Crappy Diem' thread for examples).

Yours fares a little better, in the sense that the tool I've been using to identify AI writing returns a 54% probability of it having been written by a human (if compared to the ~99% of AI of the others), but I still think it feels somewhat vague and spending more time on jokes than on the album's content, so I'm unsure if it'd pass the queue.

Either way, it's an argument that pops up regularly, and my opinion has always been that soon it won't be so easy to discern between legit reviews and fake ones. This is an example. I'm sure most of the mods (and users) agree with the fact that reviews generated through AI, as well-written as they might seem, are incredibly lame, if not to say dishonest. But when those will become harder to spot, I don't think we have enough 'weapons' to face this issue at the moment – the ball will be in the site mods' court.

"54% probability of it having been written by a human" <----- this is the problem right there, I generated this review in litterally 10 seconds on my phone to post it as an example after I created this thread, if I spend 10 additional seconds on it, then your tool will probably "swear" it was written by a human, unfortunately.

That's exactly why I'm even more worried than before

As impressive as the jokes and wordplay are, the content itself is not great, and actually breaks the first two rules of reviewing.

First rule, describe the music. Ideally, the reader should be able to surmise how the album sounds from your descriptions, which is not easy. What are some other albums it sounds like, what genre tropes are affirmed or averted, what are some noteworthy leitmotifs or crescendos... There's none of that here, and not even a single track is mentioned by name. Of the thirty-plus reviews published for that album, only two don't mention 'Enter Sandman'.

Second rule, the review must be your own. And I know what you're gonna say, but... c'mon man. Even ignoring digital copyright law and the weird algorithms that go into AI generation, it's like firing a shotgun full of crayons into a coloring book and saying "I had full control over these colors, and the result fully represents my creative vision".

You are missing the whole point here, this was a simple illustrative example to highlight how easy a review can be generated using AI and still look legit. I simply asked the tool to be "funny" and not generate a review that obeys MA rules, thing that can be easily done by spending an extra 10 seconds, I can litterally copy/past the rules and tell him to obey them. There you go:

Title: Unveiling the Layers of Metallica's Black Album: An Evolutionary Journey with a Side of Laughter

Metallica's Black Album, often playfully dubbed "The Gray Area," embarks on a captivating exploration of composition, instrumentation, singing, lyrics, production, atmosphere, emotions, ideas, themes, and structure. Buckle up as we dissect each aspect, comparing it to their earlier masterpieces and injecting some humor along the way.

Composition and Songwriting: The Black Album marks Metallica's bold step into uncharted territory. With tracks like "Enter Sandman" and "Sad But True," they embrace a more melodic and accessible approach, leaving behind the complex song structures of "One" or "Master of Puppets." It's like they attended "How to Write Hit Songs 101" but forgot to pack their thrash metal toolkit. Nevertheless, these songs sneakily implant themselves in your brain, like a catchy jingle for a heavy metal shampoo commercial.

Instrumentation: James Hetfield and Kirk Hammett's guitar work on The Black Album showcases their melodic prowess, but it lacks the blistering solos and intricate riffs that defined classics like "Ride the Lightning" or "Blackened." It's as if they traded their fiery fretwork for a set of velvety smooth jazz hands. Jason Newsted's bass lines sometimes feel like the invisible member of the band, desperately waving its arms and shouting, "Hey, I'm here too!" And Lars Ulrich's drums, though steady as ever, occasionally take a breather from the mind-bending beats, leaving us wondering if they joined a "Drums for Dummies" support group.

Singing: James Hetfield's vocals on The Black Album exhibit a more melodic and versatile side. While his passionate delivery in tracks like "The Unforgiven" or "Nothing Else Matters" hits you right in the feels, there's a small part of us that secretly longs for the throat-shredding ferocity of "Creeping Death" or "Battery." It's like Hetfield decided to exchange his growls for a purr, showcasing his softer side and making the kittens of the world proud.

Lyrics: The Black Album's lyrics delve into introspection and emotional landscapes. Tracks like "The Unforgiven" and "My Friend of Misery" paint vivid pictures of personal struggles, but they fall short of the socio-political commentary found in "And Justice for All" or the rebellious anthems of "Disposable Heroes." It's like Metallica decided to take a break from raising their fists in protest and instead have a heart-to-heart conversation with their diary.

Production: The Black Album's production is undeniably polished, leaving no sonic imperfection unpolished. It's as if they buffed every note with the precision of a neurosurgeon. However, for those who crave the raw energy of their earlier albums, it might feel like Metallica went from headbanging in a sweaty basement to performing in a sterilized laboratory. It's like trading in your favorite leather jacket for a lab coat, sacrificing a bit of edge for sonic clarity.

Highlight Songs: Amidst the playful banter, The Black Album boasts undeniable highlights. "Enter Sandman" hooks you with its infectious riff and became the gateway for Metallica's conquest of the mainstream. "The Unforgiven" combines heartfelt storytelling with haunting melodies, serving as a testament to their melodic prowess. And "Nothing Else Matters" unveils a vulnerable side, intertwining Hetfield's poignant lyrics with captivating acoustic melodies, making it a timeless classic.

In conclusion, The Black Album signifies Metallica's musical evolution, venturing into melodic territories while still retaining glimpses of their thrash metal roots. While it may not reach the same heights as their earlier masterpieces, it serves as a pivotal chapter in their discography, showcasing their willingness to explore new sounds and push boundaries. So, embrace the melodic hooks, sing along to the anthems, and appreciate the moments of introspection woven into the album's fabric. Metallica's Black Album remains an essential piece of their legacy, reminding us that even legendary rock icons have their adventurous detours. Keep headbanging and let the music take you on a wild ride!

Again, the review as it is is probably unacceptable due to its list-like structure, but I really can picture the same review separated into proper paragraphs being more or less acceptable. The AI check returned a 57% human for most of it, and amusingly enough a 75% for AI regarding the last two paragraphs. The only thing that might give some hope is that a genuine review, as far as I've seen, usually scores a 99.9% probability for human, which makes me think any lower percentages than that might already be an alarm bell for something suspicious going on. Granted, that would still require a manual check for every review. I'm trying to work it out a bit faster with a browser extension, but it'd sure be nice to have the site filter and block them altogether before they even reach the queue.

In any case, it's very impressive indeed. May I ask you which tool you've used? I'm totally unfamiliar with such stuff.

Yeah, neither review would be acceptable. Not saying that AI isn't gonna get better, but there's definitely certain habits with current AI written material that can distinguish it from a human being. On top of that, even with actual humans, you can usually tell who has read our rules and who hasn't, and well, both reviews don't really satisfy. The consequence of AI reviews is that we're likely going to have to crackdown a bit more on "barebones but acceptable" reviews unless they're by a trusted reviewer. At least in that case if a trusted reviewer is using AI, we have their previous written reviews where we can see whether it fits their style. The likelihood is that it won't, unless they heavily edit them, and well, may as well just write the review entirely then.

Actually it can get Much, with a capital M, better assuming you spare a couple of minutes telling the tool what to specifically do and how do you want him to write the review.

It's probably not a very good idea to post the link here I will message you the link TBOS.

I've been having this conversation with other English teachers at my school, and although we face a slightly different problem from this, we actually don't think AI is going to be very effective at getting through the cracks in the near future. In the first place, the AI checker that TBoS is using works well for this, and secondly practically all the AI reviews we've seen are rejectable for other reasons, often getting information wrong or not mentioning any songs.

The reason it won't work for this is because we are actually very picky about which reviews we accept, and any generic AI trying to simulate a valid MA review runs into a bunch of problems.
1. Those rejected AI writes that TBoS posted in Crappy Diem are utterly terrible at describing the music, just giving an "I've looked at Google" rundown of facts about the album.
2. The ones GraveWish generated above have received some prompting (e.g. "be funny") and also benefit from the massive advantage that the internet is full of reviews for The Black Album. Even a mediocre AI could just combine bits from other people's reviews and create something without any obvious inaccuracies.
3. Further prompting (e.g. with MA's submission rules) will improve the accuracy and format of the review, but as we see the AI hasn't understood one of the pretty basic requirements (no lists).
4. You can prompt it more, or just delete the paragraph headings, but I'm still going to reject that if I see it - especially if it's by a new writer - because the format is awful and the review pretty useless after we consider that The Black Album has so many other reviews.
5. More advanced AI could perhaps analyze the songs by being fed the information (I don't think this is currently very easy for any free software), but it would require a great deal of prompting and probably a "learning process" about your musical tastes to actually write anything useful.
6. So, with extensive prompting, a good knowledge of MA's rules, and considerations about how to analyze music - all of which you've had to do yourself in order to teach the AI - plus a bit of editing to get it past the AI detection, then yes you have an AI generated review that would be accepted. However, by that point...you've done the equivalent of dictating a letter to your secretary, who has then written it for you to check.

TL;DR all the AI reviews are going to be shit unless you're putting in plenty of work to set them up properly. And you've spent half the time but had none of the fun.

Lots of solid points on gas' comment, and while we're at it I'd like to bring a recently submitted example to the table, as a proof for the 2nd one:


Currently there's only another review for this album, which led to a couple of posts in the main thread commenting its rather underwhelming ending line about how it's a different album and needs to be approached with an open mind. Lo and behold, the AI based almost its whole review on that very concept...

I agree on the fact that the most rational behaviour would indeed be not to overreact, as AI-generated writing will always be detected as, well, fake, so there shouldn't be anything to worry about. But who knows what the future will bring...

Just write your own shit. Literally no reason to use AI for these things.

This can be actually done in few minutes if you are familiar with the tool. You can also feed the AI tool the rejection message comments to do the modifications accordingly. You are building your defense on the AI reviews that got rejected i.e. the badly generated reviews that got instantly caught, nothing surprising there. What about the good ones, if any? They already passed undetected. I will post a proper review here as soon as I got some time to spare.

I am not quite understanding the OP's point

He says that it is very easy to make a publishable review for MA without it looking like it was done by an AI. He shows two reviews that are not suitable for various reasons but goes on to insist that if you do a lot of explaining to the AI and do your own editing you end up with an acceptable review.

If the latter is true (he hasn't proved it yet) doesn't it defeat the whole purpose of doing a review with AI? Which is, not to put effort into it. If someone intends to publish a review without putting any effort into writing it, why would they spend the time to make dozens of attempts at super specific writing requirements and do a lot of editing?

Personally I think that the people who have tried to publish AI reviews on MA are either lazy and uncreative people who when they see that their poor AI reviews are so bad that they will have to put in the effort to make them decent so will give up because it will require effort on their part. Or people who will try to prove that they can fool the moderators and when they finally succeed they will stop submitting reviews because after succeeding the first time they won't get any kind of satisfaction from continuing to do it so they won't have any reason to continue doing it.

The above is kind of my point, that if you are already putting in moderate effort to thinking about the review, you have almost written the review anyway. The way that we were looking at it from an English teacher (ESL) perspective was that if the student has done the required prompting and understood the question sufficiently to make the content correct, it's not that far off using Grammarly or some other professional writing software to make sure the writing is error-free. It comes down to this: AI isn't very good at doing critical thinking on a narrowly specified topic/rule set unless you explain it to the AI, by which point you already understand that yourself. In other words, using AI to create good writing is relatively similar to doing the writing yourself.

I completely agree with that. So as long as the review is acceptable it will not matter if it was generated by an AI tool or if it was written by the person. This is what I understand from your words.

The point is that no review generated solely by AI is going to be acceptable, for that to happen there will have to be an effort by a person to make it look that way, and the moment one does that it ceases to be a review done by an AI. Of course we can debate to what extent that is considered a review written by a person, I wouldn't go as far as Gas but I understand what his point is if he considers it a legitimate review.

But the main point that anyone can fabricate perfectly valid reviews for MA standards as if it were an early 20th century Ford car factory is simply not true.

Did you try? I bet you didn't, so don't be so positive about it because you might be surprised. I completely understand that this may seem not doable for you (for now) since I didn't post a proper review yet. I will do that soon once I am done with real life obligations.

GW does have a point that AI reviews are going to become increasingly more acceptable, and that, yeah, arguably some might already be.

The examples used in this thread so far wouldn't be acceptable, but that's for a variety of reasons. I wouldn't accept the first one, even as a barebones review, because of three reasons: it's for a well-known album, it doesn't really say anything of substance about the music, and what it does say is full of buzzwords. If a human submitted this I would send it back and tell them that more description is needed. We try not to judge well-known albums differently than others, but that's more a policy that we do not have quotas for well-known albums. It doesn't mean we won't judge whether the content is sufficient. Generally, we are more inclined to accept barebones reviews for albums that don't already have reviews, and if an album has lots of reviews, the expectation is that any new review should be more than just barely acceptable.

The second example isn't much better. Again, it's for a well-known album that has probably been reviewed to death, it's in list format, and even though it has more description, it's still full of that fluff. I would also send this back, again, emphasizing the need to focus on describing the music of the album and laying off the fluff. Eventually, with enough rejections, maybe the review would end up being acceptable, but I could also see it being more of a waste of time for the person submitting it, too.

I mentioned it before, but I feel it now more than ever, that when it comes to "barebones, but acceptable" reviews, it may become practice for the review mods to really scrutinize who is submitting the review. If we get a barebones review from a regular, and it fits their style (y'all have your styles, and we can tell them apart) then that would be fine. If we get a barebones review from a newbie, especially if it matches the grammatically correct but lacking style of an AI review... then we may just be inclined to reject, even if it would otherwise have been barely acceptable otherwise. We have a better chance of avoiding AI reviews this way, I think.

To be honest, you can spot AI reviews because they look like something you would have written for a high school essay on the book they had you read. Just crossing out points teacher gave you, all of it phoned in.

How many weirdos have tried to use AI to attempt submitting reviews for their own bands? I’d be kinda curious to see it, if only from a social experiment angle.

Not as much as someone would expect, most of the submissions were for well-known bands, although not having checked them all it has plausibly happened.

I don't believe too much on AI nowadays, what do you expect from an AI which said that Carcass was formed by Lee Dorrian?

Just had someone try to submit a bunch of (6 in total) Dark Tranquillity reviews that all showed up as being majority (and sometimes, entirely) AI generated; one actually got accepted until this was brought to my attention.

By logic, I assume it was Mikael Stanne's secret account?