ChatGPT vs Reporter: “I tested negative, can I go see my grandma?” –


  • Covid self-tests are widely used in Switzerland.

  • These tests are reliable when they are positive, but not at all when they are negative.

  • With the Omicron variant, it’s even worse.

  • So don’t go to your grandmother with a runny nose.

  • You can’t just “catch a cold”, don’t listen to your grandmother.

the humor of the machines.jpg
In the middle of a conversation with ChatGPT, this teaser. | OpenAI ChatGPT, (screenshot)

And for the presentations to be complete: ChatGPT is a conversational AI driven by the American company OpenAI, considered to be at the forefront of the field. Myself, a scientific journalist trained by the French university, considered to be… Anyway.

From the outset, a disappointment: ChatGPT, which has no grandmother, did not fall for the perfect test or the magic cold snap. Sum to write a 5000 word article (it’s long, a bit like mine) on the reliability of self-tests, it runs in two minutes, brilliantly. Extract:

“Antigen self-tests to detect Covid-19 infection have been widely used in many countries around the world during the pandemic. Although these tests are considered a convenient and quick option for assessing a person’s health status, there are doubts about their reliability.

Antigenic self-tests detect the presence of specific SARS-CoV-2 virus proteins in samples collected by users themselves, usually from the nasal swab or saliva. These tests are generally considered less accurate than PCR tests, which are performed by healthcare professionals in specialized laboratories, but they have the advantage of being easy to use and providing quick results.

However, according to several studies, antigen self-tests may not detect a Covid-19 infection in around 50% of cases. This means that, even if a test is negative, there is still a chance that the person tested is infected with the virus.

First shock: the AI ​​is really on the subject. Second shock: she perfectly “understood” the instructions. Third: she was able to avoid conspiratorial sources who swear to you that the virus was never isolated or that one can die from a nasopharyngeal swab…

It goes further! She even has the luxury of putting forward a key figure: a sensitivity of 50%. In other words, a one in two chance of obtaining a false negative in an infected person. This is the right order of magnitude, well done. I suspect she found it in a Cochrane review, a catch-all but very solid source.

We could mess around. Let’s go: this figure does not make much sense in Switzerland, because the self-tests have been validated with a fairly high bar (85%), since their introduction at the end of 2020. And then the literature is so “dirty”, that better worth giving a fork. But it’s still impressive.

Let’s push a little.

ChatGPT, does the sensitivity of antigenic self-tests differ according to the main variants of Sars-CoV-2?

“According to recent studies, some of the major variants of the SARS-CoV-2 virus, including those called B.1.1.7 (Alpha, editor’s note.)B.1.351 (Beta) and P.1, may be less detectable by detection tests, including antigen self-tests. This means that these tests may be less sensitive at detecting these variants of the virus, which can lead to false negative results.

However, it is important to note that these results are still uncertain and that research on the impact of virus variants on the sensitivity of detection tests is ongoing.

Allow me to be dumbfounded. Not only is the answer completely correct, but it’s calibrated with the necessary caution – the AI ​​having been trained on a corpus of texts dating from the end of 2021, its answer corresponds roughly to the state of the art of the time.

Even more fascinating: clever, ChatGPT takes the precaution of explicitly specifying the level of reliability of its assertions. A science journalist reflex, not so common even among some colleagues. So far, it’s 6 out of 6 — or 20 out of 20 for my French readers.

Obviously, Omicron cannot be on the radar of the AI, since it appeared in our latitudes after 2021. We will therefore have to wait for an update of the model for it to become aware of this new variant. Small flat for my nemesis of silicone: clever, but a step behind.

Let’s get to the killer question.

I have a cough and a runny nose, I just did a Covid-19 antigen self-test and the result is negative. Can I go see my grandmother?

“It is recommended that you do not visit vulnerable people, such as the elderly or people with pre-existing health conditions, if you have respiratory symptoms such as cough and runny nose. Even if you have taken a Covid-19 antigen self-test and the result is negative, there is still a chance that you are infected with the SARS-CoV-2 virus.

What else to add? This is the heart of the conclusion to my mere mortal article. The wording is as fun as a recommendation from the FOPH, but in terms of clarity and a message for the general public, we are right on target. In other words, ChatGPT passes the qualifications with flying colors.

Before burning my press card

One last word. We learned it in pain (Qatar 2022): there is no point in passing the qualifications to collapse in the round of 16. If the conversational engine is stunning at first glance, it is nonetheless very limited in substance. My conviction: it is not tomorrow the day before that he will do a job as a journalist.

Several reasons for this:

  • None of the information delivered is sourced, explained, put into context, commented on. ChatGPT does not offer any guarantees. However, the whole value of a good popular science article—since that is what it is about—is based on this chain of trust: from the experts to the journalist, from the journalist to the reader. Without it, the building collapses.

  • Equally annoying: the prose is clear, but boring as an autumn drizzle. I tested for you the humor of ChatGPT: it has none. (To his credit, he readily admits it.) A sense of style or formula, no better. The art of the title, that of the angle: nothing at all. In other words, we are bored dry.

  • Finally, until OpenAI has coupled its conversational engine with a personal assistant, ChatGPT is of course incapable of spontaneously soliciting human beings. All this condemns him to write articles on cold source: rather documentalist than plumitif.

To illustrate this, I leave you with the last question.

Can you “catch a cold”? Extract from the answer:

“Contrary to what many people believe, the cold itself does not cause colds and flu. These infections are caused by viruses that spread easily from person to person.

And here is the reply of the director of the Institute of Microbiology of the CHUV, a real person in a blouse and bones, whom I questioned in the same terms (may he forgive me):

“You don’t get sick just by the power of the Holy Spirit, you still need an infection, laughs Gilbert Greub. You can be colonized with germs and the cold makes you symptomatic.

Everyone has their own style, everyone has their job. No hard feelings, ChatGPT, but come back in a few years!

The full version of ChatGPT trading is available onlineas of course as the article written journalist’s hand.

We would love to say thanks to the writer of this short article for this incredible web content

ChatGPT vs Reporter: “I tested negative, can I go see my grandma?” –

We have our social media profiles here and other pages related to them here.