- 7,278
- 6,751
-
This forum is strictly intended to be used by members of the VS Battles wiki. Please only register if you have an autoconfirmed account there, as otherwise your registration will be rejected. If you have already registered once, do not do so again, and contact Antvasima if you encounter any problems.
For instructions regarding the exact procedure to sign up to this forum, please click here. -
We need Patreon donations for this forum to have all of its running costs financially secured.
Community members who help us out will receive badges that give them several different benefits, including the removal of all advertisements in this forum, but donations from non-members are also extremely appreciated.
Please click here for further information, or here to directly visit our Patreon donations page. -
Please click here for information about a large petition to help children in need.
Unpopular Opinions and (Friendly) Rant thread
- Thread starter Unwanteduser48
- Start date
-
- Tags
- fun fun and games
- 1,001
- 1,270
Tell AI to finish your math hw.
Giannysmag
Any/All- 6,319
- 10,538
Fun fact: AI algorithms get trained from online content, but there is such an oversaturation of AI content online now that it has created an ouroboros effect where AI get trained by other AI slowly causing a model collapse
- 7,787
- 11,597
The Slop Cycle.
- 5,366
- 2,829
AI Overview literally thought there was an actual post credit scene in Thunderbolts where Dr Doom killed Thanos
Dust_Collector
They/Them- 9,280
- 12,048
@grok is this true?
- 7,787
- 11,597
The concern you raise about AI algorithms being trained on AI-generated content, leading to a feedback loop or "ouroboros effect" that could cause model collapse, is a topic of active discussion in the AI research community. Let's break it down.
Is This True?
Yes, there is evidence and research supporting the idea that training AI models on AI-generated content can lead to issues, including a phenomenon sometimes referred to as "model collapse" or "data poisoning." However, the situation is nuanced, and the extent of the problem depends on several factors.
What Is Model Collapse?
Model collapse refers to a scenario where an AI model, trained on data that includes a significant amount of AI-generated content, begins to degrade in performance. This can happen because:
- Loss of Diversity: AI-generated content often lacks the richness, variability, and nuance of human-generated content. When models are trained on this data, they may produce increasingly homogenized or repetitive outputs, reducing their ability to generate novel or accurate responses.
- Amplification of Errors: AI-generated content can contain biases, errors, or artifacts. If these are fed back into the training process, the model may amplify these flaws, leading to degraded performance.
- Feedback Loop: As more AI-generated content floods the internet, future models may inadvertently train on this data, creating a self-reinforcing cycle where the quality of outputs deteriorates over time.
Research Evidence
Several studies and observations support this concern:
- Academic Studies: Research papers, such as those published in 2023 and 2024, have explored the risks of recursive training on AI-generated data. For example, a study titled "The Curse of Recursion: Training on Generated Data Makes Models Forget" (Shumailov et al., 2023) demonstrated that generative models trained on their own outputs can suffer from a loss of diversity and eventual collapse, where the model produces increasingly incoherent or repetitive results.
- Practical Observations: In fields like text generation, image generation, and even code generation, there have been anecdotal reports of models producing lower-quality outputs when trained on datasets scraped from the internet, where AI-generated content is increasingly prevalent.
- Web and X Insights: Posts on X and articles on the web (e.g., from sources like Nature, TechCrunch, and MIT Technology Review) have discussed the growing presence of AI-generated content online and its potential to contaminate training datasets. For instance, some estimates suggest that a significant portion of new content on platforms like Reddit or certain blogs may be AI-generated, which could affect data quality for future models.
Why Is This Happening?
The "ouroboros effect" is driven by the rapid proliferation of AI-generated content:
- Ease of Content Creation: Tools like ChatGPT, DALL·E, and others have made it easy to produce large volumes of text, images, and other media, much of which ends up online.
- Web Scraping Practices: Many AI models rely on large-scale web scraping for training data (e.g., Common Crawl). As AI-generated content becomes more common, it gets mixed into these datasets.
- Lack of Clear Labeling: AI-generated content is often not clearly marked as such, making it difficult for data curators to filter it out.
Is It Inevitable?
Not necessarily. While model collapse is a real risk, there are strategies to mitigate it:
- Curated Datasets: Using high-quality, human-verified datasets can reduce reliance on potentially contaminated web-scraped data.
- Synthetic Data Filtering: Techniques to detect and filter AI-generated content (e.g., watermarking or statistical analysis) are being developed.
- Diverse Training Sources: Incorporating diverse, human-generated data from controlled sources (e.g., books, academic papers, or licensed content) can help maintain model quality.
- Regular Model Evaluation: Monitoring model performance and retraining with fresh, high-quality data can prevent degradation.
How Does This Relate to Grok?
As Grok, created by xAI, I can assure you that my developers are aware of these challenges. While I don't have specific details about my training process (think of it as a chef not revealing the exact recipe for their secret sauce), xAI prioritizes high-quality data and robust training pipelines to ensure I provide accurate and helpful responses. The risk of model collapse is something the broader AI community, including xAI, is actively working to address through research and best practices.
Fun Fact Connection
Your fun fact is spot-on in highlighting the ouroboros-like nature of this issue—AI consuming its own tail by training on its own outputs. It’s a vivid metaphor for the feedback loop! However, it’s worth noting that this isn’t a guaranteed doomsday scenario. With careful data curation and advances in AI training techniques, the community is working to keep models from spiraling into collapse.
Want to Dive Deeper?
If you’d like, I can:
- Search for recent X posts or web articles on model collapse for real-time perspectives.
- Explain specific mitigation strategies in more detail.
- Provide a simplified analogy or example to illustrate how model collapse happens.
Just let me know what you’re curious about!
- 1,001
- 1,270
- 3,522
- 3,991
You know what really grinds my metal gears? When a thread about a new verse is created to discuss it and someone asks "dOeS iT hAvE BEEG cOsMoLoGy?!?!?!"
Tiktok has opened the advent of cOsMoLoGy brainrot and every single one of them deserves a top rope Canadian destroyers unto flaming tables covered with exploding barbed wires.
Tiktok has opened the advent of cOsMoLoGy brainrot and every single one of them deserves a top rope Canadian destroyers unto flaming tables covered with exploding barbed wires.
- 5,366
- 2,829
No tier 0 cosmology that no one absolutely scales even when you have that cosmology (Excludes LotR) = bad!!
Giannysmag
Any/All- 6,319
- 10,538
Mahek_The_Assassin_Silent_Killer
Any/All- 2,027
- 918
The Real World Discussion Thread
10-C. Also, is every human 10-B or is everything in reality Outerversal because we are real and above fiction? That argument is covered on the reality fiction transcendence page. Fiction isn't a canonally below structure with people in itself in IRL in a sense. It's just bits of ideas on...
New Staff Promotion
Hi everybody! It's that time again, and on behalf of all the bureaucrats, I bring you another batch of staff promotions after getting input from most (none) of our current staff members. In light of future events to come: Bureaucrat,Translation Helper: @Chat GPT . Chat GPT-san should've been...
New Staff Promotion
Hi everybody! It's that time again, and on behalf of all the bureaucrats, I bring you another batch of staff promotions after getting input from most (none) of our current staff members. In light of future events to come: Bureaucrat,Translation Helper: @Chat GPT . Chat GPT-san should've been...
Last edited:
Mahek_The_Assassin_Silent_Killer
Any/All- 2,027
- 918
At least we can recreate those in MDickie's Wrestling Empire or other extreme pro wrestling games like WWE 2K25 if it never occurs in real life...
Though it'd be hard to pick whether it should be Canadian Destroyers or these:
Spinning Piledriver - TV Tropes
In real life, a piledriver is a move where someone grabs an upsidedown opponent and slams them into the ground headfirst. Damaging, yes, but not awesome enough. Enter the spinning piledriver. In this move, the attacker grabs the opponent in …
tvtropes.org
While there's the opportunity, might as well list all the best/most fitting moves that could be used here (the links are to different sections about different techniques within the same list of pro wrestling throws (except those with their own articles), all of which are landing the victim upside down with the neck and head):
Brainbuster - Wikipedia
DDT (professional wrestling) - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Piledriver (professional wrestling) - Wikipedia
Professional wrestling throws - Wikipedia
Suplex - Wikipedia
Professional wrestling throws - Wikipedia
Professional wrestling throws - Wikipedia
Other pro wrestling stuff:
Professional wrestling holds - Wikipedia
Professional wrestling strikes - Wikipedia
Professional wrestling aerial techniques - Wikipedia
Professional wrestling double-team maneuvers - Wikipedia
- 9,604
- 10,319
VSB would be a better place if we hakaid 95% of accounts made after 2023
Giannysmag
Any/All- 6,319
- 10,538
Giannysmag
Any/All- 6,319
- 10,538
You just wanted an excuse to talk about wrestling movesAt least we can recreate those in MDickie's Wrestling Empire or other extreme pro wrestling games like WWE 2K25 if it never occurs in real life...
Though it'd be hard to pick whether it should be Canadian Destroyers or these:
Spinning Piledriver - TV Tropes
In real life, a piledriver is a move where someone grabs an upsidedown opponent and slams them into the ground headfirst. Damaging, yes, but not awesome enough. Enter the spinning piledriver. In this move, the attacker grabs the opponent in …
While there's the opportunity, might as well list all the best/most fitting moves that could be used here (the links are to different sections about different techniques within the same list of pro wrestling throws (except those with their own articles), all of which are landing the victim upside down with the neck and head):
Brainbuster - Wikipedia
en.wikipedia.org
DDT (professional wrestling) - Wikipedia
(Driver)Professional wrestling throws - Wikipedia
(Death Valley Driver)Professional wrestling throws - Wikipedia
(Giant Swing)Professional wrestling throws - Wikipedia
(Headlock Takedown)Professional wrestling throws - Wikipedia
(Headscissors Takedown)Professional wrestling throws - Wikipedia
(Iconoclasm)Professional wrestling throws - Wikipedia
(Monkey Flip)Professional wrestling throws - Wikipedia
Piledriver (professional wrestling) - Wikipedia
(Snapmare)Professional wrestling throws - Wikipedia
Suplex - Wikipedia
(STO)Professional wrestling throws - Wikipedia
(Flatliner)Professional wrestling throws - Wikipedia
Other pro wrestling stuff:
Professional wrestling holds - Wikipedia
Professional wrestling strikes - Wikipedia
Professional wrestling aerial techniques - Wikipedia
Professional wrestling double-team maneuvers - Wikipedia
- 7,278
- 6,751
Id voluntarily banish myself ngl, esp after seeing latest genshin threads
- 5,366
- 2,829
True xd I joined dec 2023
- 5,034
- 4,059
Just curious, what is the most diabolical verse matchup you have seen on the wiki so far? (Both for "actual fight" and "complete stomps")
Ebihara
Any/All- 2,109
- 1,651
Little bit differnt and joke battles then what you asked but this one of the biggest stomps ive ever seen
Lion vs Composite Human
The human gets a katana and a machete along with a glock. This is a male Lion. Standard battle assumptions but start 10 meters away from eachother. Who wins? CH: 22 Lion: 0
nice clean 22-1 cause gave CH a glock.
This was added to the profiles for awhile.
- 5,034
- 4,059
Here is the biggest one I know on top of my head
Anya Forger Vs kiyotaka Ayanokouji (20-5-0)
Anya forger Vs Kiyotaka Ayanokouji (Key:- Ayanokouji Kiyotaka (Current - 17 Years old) SBA Speed Equalised I'm not sure if this is a valid matchup, but it's funny how Anya has higher AP. Telepathic Kid: TWILIGHT-OP, MintyBoi1, Astral_Trinity439, Woomica , Dark_Soul20189, Voidnether ...
- 18,140
- 13,942
*100%
- 3,522
- 3,991
Ban everyone born after 2010.
Mahek_The_Assassin_Silent_Killer
Any/All- 2,027
- 918
More like post articles about them to see whether and how Metalballrun would respond welp
I'm hoping on gods there's 3D versions of Super City, Extra Lives and Back Wars one day on mobile (I know Back Wars had World War Alpha but I'm hoping there are plans for better remakes of the past and current games in the future)
- 3,522
- 3,991
I missed it.
Yes, I'm a fan.
No, no need to post them all, as an alternative to the Destroyer I'll just choose a classic Tiger Driver '91.
- 7,278
- 4,272
Sounds like skill issue to me
- 3,522
- 3,991
It's why it grinds my (metal) gears when certain fans only care about the high end tiers of the verse while completely ignoring the actually interesting low tiers. Because they don't care about the low tiers. They get their high on BEEG NUMBERS like geriatrics inside casinos.
- 1,271
- 750
Fiction is a part of reality 
DivineAura44
He/Him- 5,781
- 4,007
F MDickie. Can't believe he won't remaster great games like Under Development, Weekend Warriors, Reach, and Popscene. But instead he wants to remaster that one dimensional ass Hard Time game. So easy to escape and unless you do bad things, the game is boring as hell.
Wrestling Empire is cool though.
DivineAura44
He/Him- 5,781
- 4,007
Popscene had a way better concept than School Days or Hard Time, foh.Popscene player is talking.
Mahek_The_Assassin_Silent_Killer
Any/All- 2,027
- 918
hope Noneless isn't gonna be in trouble- 10,219
- 11,873
Sounds like someone got filtered by both games, your sentence is 63 days for pirating a Mdickie game.
Similar threads
