OpenAI hallucinates 37% of the time. The Washington Post is changing its tune. And giving up on the Canada mass grave hoax.
OpenAI Hallucinates 37% of the Time
There is so much hype about AI, ranging from the utopian (it will usher in a new era of knowledge and prosperity) to the apocalyptic (it will attain consciousness and wipe out humanity).
One problem, though, for either expectation: Like a paranoid hippie on LSD, AI keeps hallucinating. That is the term used for making up an answer and presenting it as true.
According to an article in Futurism, OpenAI’s new large language model, GPT 4.5, hallucinates 37% of the time. But the $300 billion company is presenting this as a good thing. Its previous model hallucinated 61.8% of the time. And the model before that hallucinated 80.3% of the time.
That is progress, so maybe future models will sober up. And the numbers come from a sophisticated testing program. The easy questions you and I might ask ChatGPT probably show a better record.
Still, as I’ve recounted at this blog, my attempts to use it for my research have come up with some howlers, such as confusing Martin Luther with the Roman poet Seneca and listing as the best book on a topic a title that doesn’t exist by an author who also doesn’t exist. Recently, I came across two female writers with the same last name and I was curious if they were related. I asked AI, which informed me that the two have the same last name because they are both married to the same woman! A simple Google search confirmed that they are, in fact, sisters.
Most hallucinations, though, are less obvious. The problem is that when we are using AI to get information, the findings are presented as factual, whether they are or not. So people who use that information–such as doctors, engineers, scientists, journalists, etc.–may not realize that the information is bogus, unless they check it the old fashioned way.
Artificial Intelligence is not intelligent enough to say, “I don’t know.”
The Washington Post Is Changing Its Tune
America’s journalism profession is in disarray, as newspapers can’t compete with the internet, the left-wing bias of the news media was thwarted by Trump’s election, and only 31% of the public trusts what they hear in the news.
So now, some owners of the news media are trying to change their product to make it more palatable to the general public.
The most notable shakeup is happening at the Washington Post, the newspaper of record for our nation’s capital that has long been a bastion of liberal journalism. Jeff Bezos, the founder of Amazon who owns the Post, recently announced a big change in the opinion pages:
“We are going to be writing every day in support and defense of two pillars: personal liberties and free markets,” he wrote. “We’ll cover other topics too of course, but viewpoints opposing those pillars will be left to be published by others.”
Personal liberties! Free markets! That sounds right wing to progressive reporters! The announcement led to a spate of resignations and subscription cancellations. Also the charge of “fascism,” as if fascists support personal liberties and free markets. Also charges of “oligarchy,” that billionaires are trying to control the news and that Bezos is just another tech-bro who is groveling before Donald Trump.
Similar changes are being made on the other coast, with another newspaper that has served as a liberal mouthpiece, the Los Angeles Times. Biotech billionaire Patrick Soon-Shiong, who like Bezos forbade his paper from endorsing Kamala Harris, is approaching making the adjustments in a characteristically California, high-tech way.
As reported in the UK Guardian,
Some Los Angeles Times opinion pieces will now be published with an artificial intelligence-generated rating of their political content, and an AI-generated list of alternative political views on that issue, the paper’s biotech billionaire owner announced on Monday.
The new AI “Insights” feature will only be applied to a range of opinion content in the paper, not its news reporting, according to a public letter announcing the change from Patrick Soon-Shiong, the medical entrepreneur who bought the Los Angeles Times in 2018.
The AI-generated tool “operates independently” from the paper’s human journalists, and “the AI content is not reviewed by journalists before it is published”, the Los Angeles Times noted in a summary of the new feature.
That is likely to just confirm and prove the left-wing bias, except perhaps for hallucinations. And I’m thinking that the Bezos reforms won’t work very well either. The markets in both D.C. and L.A. are very liberal, so trying to win over the larger vibe-shifting nation as a whole probably isn’t a winning strategy. And the 69% of Americans who don’t trust the media probably won’t trust these changes either.
Giving Up on the Canada Mass Grave Hoax
You may remember that in 2021, reports came out in Canada about radar technology discovering over 200 human remains in unmarked graves surrounding a Catholic boarding school for Indians.
This led to an overwhelming revulsion in the media about attempts to assimilate Native American children turning into mass genocide. Speculation that thousands of children may have been buried in mass graves led to headlines like this one in the New York Times: “‘Horrible History’: Mass Grave of Indigenous Children Reported in Canada.” Activists protested, Pope Francis came to Canada to apologize, 24 churches were set on fire, and Prime Minister Trudeau said the arsons were “fully understandable given the shameful history we are all becoming more aware of.”
“There’s one problem,” reports National Review. “In all the excavations that have been conducted, not a single unmarked grave has been found. Not one.”
Finally and quietly, after years of futile investigations and excavations, the government commission tasked to expose the alleged atrocities was shut down.
Ad-Free & Subscriber Log-In