Kingdom

United

when

Large language models (LLMs) trained on previous iterations of AI-generated material produce outputs that lack substance and nuance, a new study has found. The findings present a new challenge for AI developers, who rely on limited human-generated data sets for content. 
Also read: <a href="https://www.cryptopolitan.com/ai-deepfakes-making-it-hard-for-us/">AI deepfakes are making it hard for US authorities to protect children – report</a> 
Artificial intelligence researchers from the University of Cambridge and Oxford University in the United Kingdom tried to write prompts relying on a dataset comprising only AI-generated content.&nbsp;The outcome was not ideal, as it produced incomprehensible responses. 
<h3 id="ai-still-needs-humans-to-make-sense">AI still needs humans to make sense</h3> 
One of the paper’s authors, Zhakar Shumaylov from the University of Cambridge <a href="https://www.nature.com/articles/d41586-024-02420-7#:~:text=Collapse%20happens%20because%20each%20model,ones%20being%20regurgitated%20is%20boosted." rel="nofollow noopener" target="_blank">said</a> there is a need for quality control in the data that feeds LLMs, the technology behind generative AI chatbots like ChatGPT and Google’s Gemini. Shumaylov said: 
<blockquote>
 “The message is we have to be very careful about what ends up in our training data. [Otherwise,] things will always, provably, go wrong”.
</blockquote> 
The phenomenon is known as “model collapse,” Shumaylov detailed. It has been proven to affect all kinds of artificial intelligence models including those that specialize in image generation using text prompts. 
According to the <a href="https://doi.org/10.48550/arXiv.2404.01413" rel="nofollow noopener" target="_blank">study</a>, repeat text prompts using AI-generated data on one model ended up generating gibberish. For example, researchers found that one system tested with text about the UK’s medieval Church towers produced a repetitive list of jackrabbits after only nine generations. 
Commenting on the outputs, University of California computer scientist, Hany Farid, likened the data collapse to the challenges endemic to animal in-breeding. 
“If a species inbreeds with their own offspring and doesn’t diversify their gene pool, it can lead to a collapse of the species,” Farid said. 
When the researchers infused human-generated data into the AI data, the collapse happened more slowly than when it was running on purely AI-generated content. 
<figure id="attachment_536559" aria-describedby="caption-attachment-536559" style="width: 100%" class="wp-caption aligncenter">
 <img fetchpriority="high" decoding="async" class="wp-image-536559 size-full" src="https://www.cryptopolitan.com/wp-content/uploads/2024/06/istockphoto-2153983313-612x612-1.jpg" alt="AI" width="612" height="408" srcset="https://www.cryptopolitan.com/wp-content/uploads/2024/06/istockphoto-2153983313-612x612-1.jpg 612w, https://www.cryptopolitan.com/wp-content/uploads/2024/06/istockphoto-2153983313-612x612-1-300x200.jpg 300w" sizes="(max-width: 612px) 100vw, 612px">
 <figcaption id="caption-attachment-536559" class="wp-caption-text">
 AI models produce gibberish when trained on AI-generated data.
 </figcaption>
</figure> 
<h3 id="researchers-ai-could-worsen-biases-against-minority-groups">Researchers: AI could worsen biases against minority groups</h3> 
Language models work by building up associations between tokens — words or word parts — in huge swathes of text, often scraped from the Internet. They generate text by spitting out the statistically most probable next word, based on these learned patterns. 
Also read: <a href="https://www.cryptopolitan.com/ai-powered-teddy-bear-poe-reads-stories/">ChatGPT-powered teddy bear Poe reads bedtime stories to kids</a> 
The study, which was published in the journal Nature on July 24, showed that information mentioned a few times in data sets is likely not to be repeated. Researchers worry this could negatively impact already marginalized minority groups. 
To avert the model collapse in real-life use cases, the study suggested watermarking AI-generated content and human-generated content. But this could also be problematic due to a lack of coordination between rival AI companies, it said. 
The study’s findings come at a time when there is increased debate on whether AI will result in the total exclusion of humans in the creation of content, including the writing of novels and newspaper articles. 
Titled, ‘AI models collapse when trained on recursively generated data’, the study’s outcomes put that debate to rest – humans aren’t being removed from the equation just yet.

AI models spew ‘gibberish’ when trained on AI-generated data

目錄

Cryptopolitan_News

在 2024年7月26日發表的其他文章

XRP Poised for Parabolic Surge? Weekly Bullish Candle Ignites 800% Rally Speculation

Foreign crypto miners could be spying on US bases: Elizabeth Warren

SLERF Bounces Back Over 50% This Week: Will The Momentum Last

Solana Memecoin Trump Fired Biden Will Rally 17,000% Before Its First Exchange Listing, As SHIB a...

DeepMind AI hits silver-medal standard in International Mathematical Olympiad

Bitcoin Price Recovers Lost Ground: Is the Bull Run Back?

訂閱免費電子報，每天收取加密貨幣最新動態！