Experts sound alarm as troubling results surface from latest ChatGPT model tests: 'Untrustworthy at best'

AI has already begun eating its own tail, according to the Guardian. ChatGPT has been found to be citing a highly dubious source that itself is also powered by artificial intelligence. 

What's happening?

Multiple ChatGPT queries cited Grokipedia to back up its statements. Grokipedia is a Wikipedia competitor made by Elon Musk's company xAI. Instead of relying on people to edit articles, Grokipedia articles are curated by Grok, an artificial intelligence baked into the social media platform X.  

While Grok is generally competent at producing accurate results quickly more often than not, many of the statements cited from Grokipedia have been found to be falsehoods. There is growing concern that this misinformation is a concerted attempt to erode AI quality. 

Disinformation researcher Nina Jankowicz said ChatGPT was "relying on sources that are untrustworthy at best, poorly sourced, and deliberate disinformation at worst," according to the Guardian. "They might say, 'oh, ChatGPT is citing it, these models are citing it, it must be a decent source, surely they've vetted it' — and they might go there and look for news about Ukraine."

One study suggested that 75% of data being fed into training large language models is synthetic, that is, made by other AI. This has the potential to erode the quality of AI output as the models devolve. 

Why is AI disinformation concerning?

The social implications of handing the reins of important tasks over to AI are massive. It has the potential to displace large parts of the workforce, as it is a cheaper alternative to humans.

Additionally, the quality of AI output appears to be peaking and largely remains worse than what humans can produce. If AI models are trained on bad data, they may continue repeating falsehoods even if the source material has been cleaned or removed.  

To top it all off, AI requires an incredible amount of electricity to keep working. These grid energy sources are often dirty sources, such as coal or gas, adding an environmental cost to the use of artificial intelligence. 

Tech giants such as Meta, Microsoft, and Google are trying to mitigate these costs by investing in nuclear power. Improvements in the technology may also help to taper these steep energy needs. 

What's being done about AI reliability?

When tempered by human cross-checking, AI has the potential to be useful for applications such as weather prediction, disease forecasting, and battery design. 

That said, there are plenty of other cases where help from AI has been shown to slow workers down. If this becomes a trend, employers may be less likely to push AI use.  

What's the most you'd pay per month to put solar panels on your roof if there was no down payment? $200 or more 💰 $100 💸 $30 💵 I'd only do it if someone else paid for it 😎 Click your choice to see results and speak your mind.

Get TCD's free newsletters for easy tips to save more, waste less, and make smarter choices — and earn up to $5,000 toward clean upgrades in TCD's exclusive Rewards Club.