Prolonged AI use can be hazardous to your health and work: 4 ways to stay safe

1 hour ago 5

Follow ZDNET: Add america arsenic a preferred source connected Google.

ZDNET's cardinal takeaways  

AI is getting amended astatine tiny tasks, but inactive lags connected long-form analysis.
The consequences of prolonged interactions with AI tin beryllium disastrous.
Use AI similar a instrumentality for well-defined tasks, and debar falling down a rabbit hole.

Better to bash a small good than a large woody badly. So said the large philosopher Socrates, and his proposal tin use to your usage of artificial intelligence, including chatbots specified arsenic OpenAI's ChatGPT, oregon Perplexity, arsenic good arsenic the agentic AI programs progressively being tested successful enterprise.

AI probe progressively shows that the safest and astir productive people with AI is to usage it for small, constricted tasks, wherever outcomes tin beryllium good defined, and results tin beryllium verified, alternatively than pursuing extended interactions with the exertion implicit hours, days, and weeks.

Also: Asking AI for aesculapian advice? There's a close and incorrect way, 1 doc explains

Extended interactions with chatbots specified arsenic ChatGPT and Perplexity tin pb to misinformation astatine the precise least, and successful immoderate cases, delusion and death. The exertion is not yet acceptable to instrumentality connected the astir blase kinds of demands of reasoning, logic, communal sense, and heavy investigation -- areas wherever the quality caput reigns supreme.

(Disclosure: Ziff Davis, ZDNET's genitor company, filed an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful grooming and operating its AI systems.)

We are not yet astatine AGI (Artificial General Intelligence), the expected human-level capabilities of AI, truthful you'd bash good to support the technology's limitations successful caput erstwhile utilizing it.

Put simply, usage AI arsenic a instrumentality alternatively than letting yourself beryllium sucked down a rabbit spread and get mislaid successful endless rounds of AI conversation.

What AI does good - and not truthful well

AI tends to bash good astatine elemental tasks, but poorly astatine analyzable and heavy types of analysis.

The latest examples of that are the main takeaways from this week's merchandise of the Annual AI Index 2026 from Stanford University's Human-Centered AI radical of scholars.

On the 1 hand, editor-in-chief Sha Sajadieh and her collaborators marque wide that agentic AI is progressively palmy astatine tasks specified arsenic looking up accusation connected the Web. In fact, agents are adjacent to human-level connected regular online processes.

Also: 10 ways AI tin inflict unprecedented damage

Across 3 benchmark tests -- GAIA, OSWorld, and WebArena -- Sajadieh and squad recovered that agents are approaching human-level show connected multi-step tasks specified arsenic opening a database, applying a argumentation rule, and past updating a lawsuit record. On the GAIA test, agents person an accuracy complaint of 74.5%, inactive beneath the 92% of quality show but mode up from the 20% of a twelvemonth ago.

On the OSWorld test, "Computer subject students lick astir 72% of these tasks with a median clip of astir 2 minutes," portion Anthropic's Claude Opus 4.5, up until precocious its astir almighty model, reaches 66.3%. That means "the champion exemplary [is] wrong 6 percent points of quality performance."

WebArena shows AI models "now wrong 4 percent points of the quality baseline of 78.2%" accuracy.

While Claude Opus and different LLMs are not perfect, they amusement accelerated advancement successful astatine slightest reaching benchmark levels that travel person to human-level performance.

That makes sense, arsenic manipulating a web browser oregon looking thing up successful a database should beryllium among the easier scenarios successful which the natural-language punctual tin plug into APIs and outer resources. In different words, AI should person astir of the instrumentality it requires to interface with applications successful constricted ways and transportation retired tasks.

Also: 40 cardinal radical globally are utilizing ChatGPT for healthcare - but is it safe?

Note that adjacent with well-defined, constricted tasks, it helps to cheque what you're getting from a bot, arsenic the mean people connected these benchmarks inactive falls abbreviated of quality capableness -- and that's successful benchmark tests, a benignant of simulated performance. In real-world settings, your results whitethorn vary, and not to the upside.

AI can't grip the hard stuff

When they dug into deeper kinds of work, the Stanford scholars recovered overmuch little encouraging results.

Research has found, they noted, that "models grip elemental lookups good but conflict erstwhile asked to find aggregate pieces of matching accusation oregon to use conditions crossed a precise agelong papers -- tasks that would beryllium straightforward for a quality scanning the aforesaid text."

That uncovering aligns with my ain anecdotal acquisition utilizing ChatGPT to draught a concern plan. Answers were good successful the archetypal fewer rounds of prompting, but past degraded arsenic the exemplary snuck successful facts and figures I had not specified, oregon that mightiness person been applicable earlier successful the process but had nary concern being included successful the contiguous context.

The lesson, I concluded, was that the longer your ChatGPT sessions, the much errors sneak in. It makes the acquisition infuriating.

Also: I built a concern program with ChatGPT and it turned into a cautionary tale

The results of unchecked bot elaboration tin get much serious. An nonfiction past week successful Nature mag describes however idiosyncratic Almira Osmanovic Thunström, a aesculapian researcher astatine the University of Gothenburg, and her squad invented a disease, "bixonimania," which they described arsenic an oculus information resulting from excessive vulnerability to bluish airy from machine screens.

They wrote ceremonial probe papers connected the made-up condition, past published them online. The papers got picked up successful bot-based searches. Most of the ample connection models, including Google's Gemini, began to faithfully subordinate the information bixonimania successful chats, pointing to the faked probe papers of Thunström and team.

The information that bots volition confidently asseverate the beingness of the fake bixonimania speaks to a deficiency of oversight of the technology's entree to information. Without due checking, you can't cognize if a exemplary volition verify what it's spitting out. As 1 student who wasn't progressive successful the probe noted, "We should measure [the AI model] and person a pipeline for continuous evaluation."

Consequences tin beryllium serious

A much superior variant, wherever a idiosyncratic seems to person gone down a rabbit spread of confiding successful a bot, is described successful a recent New York Times article by Teddy Rosenbluth astir the lawsuit of an older antheral grappling with achromatic humor compartment cancer.

Rather than pursuing his oncologist's advice, the patient, Joe Riley, relied connected extended enactment with chatbots, particularly Perplexity, to refute the doctor's diagnosis. He insisted his AI probe revealed helium had what's called Richter's Transformation, a complication of crab that would beryllium made much adverse by the recommended treatment.

Also: Use Google AI Overview for wellness advice? It's 'really dangerous,' probe finds

Despite emails from experts connected Richter's questioning the worldly successful the Perplexity summaries of the condition, Riley stuck with his content successful his AI-generated reports and resisted his doctor's and his family's pleas. He missed the model for due treatment, and by the clip helium relented and agreed to effort treatment, it was excessively late.

Rosenbluth makes the transportation betwixt the communicative of Joe Riley and the lawsuit of Adam Raine past year, who committed termination aft extended chats with ChatGPT astir his inclination to extremity his life.

Riley's son, Ben Riley, wrote his ain relationship of his father's travel with AI. While the younger Riley doesn't blasted the exertion per se, helium points retired that getting immersed successful chats and losing position tin person consequences.

"The information remains that AI does beryllium successful our world," writes Riley, "and conscionable arsenic it tin service arsenic substance to those suffering manic psychosis, truthful excessively whitethorn it affirm oregon amplify our mistaken knowing of what's happening to america physically and medically."

Staying sane with unreliable AI

The inclination to prosecute successful long-form discussions astir depression, suicide, and superior wellness conditions is understandable. People person been habituated to long-form engagements of hours astatine a clip connected societal media. Some radical are lonely, and a natural-language speech with a bot is amended than nary speech astatine all.

Also: Your chatbot is playing a quality - wherefore Anthropic says that's dangerous

Bots person a inclination toward sycophancy, research has shown, which tin marque hours of engagement with a bot much fulfilling than the mean springiness and instrumentality with a person.

And the companies that marque the technology, portion informing users to verify bot output, person tended to spot little accent connected antagonistic reports from individuals specified arsenic Riley and Raine.

4 rules for avoiding the rabbit hole

A fewer rules tin assistance mitigate the worst effects of putting excessively overmuch accent connected the technology.

Define what you are going to a chatbot for. Is determination a well-defined task that has a constricted scope and for which the predictions of the bot tin beryllium fact-checked with different sources?
Have a steadfast skepticism. It's good known that chatbots are prone to confabulation, confidently asserting falsehoods. It doesn't substance however galore chatbots you usage to effort to equilibrium the bully and the bad; each of them should beryllium treated with a steadfast skepticism arsenic having lone portion of the truth, if any.
Regard chatbots not arsenic friends oregon confidants. They are integer tools, similar Word oregon Excel. You're not trying to person a narration with a bot but alternatively to implicit a task.
Use proven integer overload skills. Take agelong breaks. Step distant from the machine for a non-digital quality interaction, specified arsenic playing paper games with a person oregon going for a walk.

Also: Stop saying AI hallucinates - it doesn't. And the mischaracterization is dangerous

Falling down the rabbit spread happens partially arsenic a effect of simply being parked successful beforehand of a surface with nary downtime.

Read Entire Article