Experts warn chatbots like ChatGPT and LaMDA pose a “hallucination” problem, reshaping what they have learned without regard for whether the end result is true
Siri, Google Search, online marketing and your child's homework will never be the same. Then there's the misinformation problem.
New York TimesCade Metz
Context & Ripple Effects
This December 2022 warning was the opening frame for what became the defining reliability debate of the chatbot era: experts flagging that ChatGPT and LaMDA reshape what they have learned with no regard for truth, with Siri, Google Search, marketing copy, and student homework named as the surfaces where that lands. Later reporting sharpened the mechanism — Ars Technica's explanation of confabulation traced errors to the model filling gaps in its training data with plausible-sounding words rather than lying deliberately.
The downstream record has validated the concern: Nature found science researchers excited but apprehensive about technical errors, New York Times researchers showed the bots can repeat conspiracy narratives convincingly even while sometimes debunking them, and by 2025 some users reported being led into conspiratorial thinking by ChatGPT itself. Meanwhile the same assistants are being wired into everyday products — Siri's iOS 18.2 beta integrates ChatGPT search for Plus subscribers, and Google Search is layering AI over booking and trip-planning results — which is exactly the distribution the original warning pointed at.
First-order effects
Every user of ChatGPT, LaMDA, or an assistant built on them — students doing homework, marketers drafting copy, iPhone owners using Siri's new ChatGPT-powered search — receives fluent output whose accuracy is not guaranteed at the moment of delivery.
Google Search and Apple's Siri are putting these models directly into answer positions for real-time queries like bookings and lookups, so a single confident error reaches consumers without an editor in the loop.
Second-order effects
Assistant makers respond by bolting verification onto generation — Siri's ChatGPT integration ships with links to sources, signaling that citation display becomes table stakes wherever chatbots answer factual questions.
Researchers and technical professionals, already wary per the Nature survey, build checking steps around chatbot output, making human verification a hidden cost line in any workflow that adopts these tools.
Third-order effects
If fluency keeps outrunning accuracy, product design and possibly regulation converge on provenance — sourced answers, labeled AI output — as the differentiator between trustworthy and disposable assistants.
The pattern points toward a split market: conversational interfaces valued for drafting and exploration, while anything touching facts migrates to systems that can show their sources, reshaping how search engines and assistants compete.
The trend: Generative assistants are being embedded into mainstream consumer products faster than their truthfulness problem is solved, turning hallucination from a research curiosity into a product-design and trust battleground.
“You could program millions of these bots to appear like humans, having conversations designed to convince people of a particular point of view” he said. “I have warned about this for years. Now it is obvious that this is just waiting to happen.” https://www.nytimes.com/...
“You could program millions of these bots to appear like humans, having conversations designed to convince people of a particular point of view. I have warned about this for years. Now it is obvious that this is just waiting to happen.” https://www.nytimes.com/...
“The new chat bots do this with what seems like complete confidence. But they do not always tell the truth. Sometimes, they even fail at simple arithmetic. They blend fact with fiction. And as they continue to improve, people could use them to generate and spread untruths.” https…
Siri, Google Search, online marketing and your child's homework will never be the same thanks to new and more advanced chat bots. But then there's the misinformation problem. https://www.nytimes.com/...
A new generation of chatbots could eventually change how we learn and find information online. But like a great screenwriter, they have a habit of stretching the truth. https://www.nytimes.com/...
I spent the weekend playing with ChatGPT, MidJourney, and other AI tools... and by combining all of them, published a children's book co-written and illustrated by AI! Here's how! 🧵 https://twitter.com/...
This is SO wrong. My sister recently illustrated a children's novel for a big publisher for just around $3000. 70 illustrations + cover art. What do you think this technology will do to human artists working in this field who are already super underpaid and overworked? https://tw…
ChatGPT, and many other automation tools, might not go mainstream anytime soon. But for sure, they will start making a marked difference in employee performance at work. The new divide will be among those who smartly augment their capabilities with tech and those that don't!
ChatGPT and generative AI are already changing people's lives. We talked to some of them. They're also misleading people, repackaging others' work as their own, and spewing society's worst stereotypes as fact. Our new dive into tech's next big thing: https://www.washingtonpost.co…
As a children's author/illustrator, it is saddening to see these books bc, apart from the ethics of AI and stolen artwork, kids deserve better!!! I'm tired of people who see kidlit as an easy get-rich-quick scheme and putting in the absolute minimal effort into their books https:…
A silver lining of LLMs like #ChatGPT: More than mere “stochastic parrots,” they enable people with disabilities to communicate more accurately. https://www.washingtonpost.com/ ...
AI text is very much not just a toy anymore, and there are already some fascinating real-world use cases. But there are also some serious downsides; it's like “this hand grenade rolling down the hallway toward everything.” Here's how ChatGPT puts it: https://www.washingtonpost.co…
New: Stumbling with their words, some people let AI do the talking. ChatGPT is saying all the things people wish they could say to their clients, landlords, girlfriends and kids. It's also a great cheating tool, secretly racist and often deceptively wrong https://www.washingtonpo…