Reporter Holly Else in a news article for Nature:

An artificial-intelligence (AI) chatbot can write such convincing fake research-paper abstracts that scientists are often unable to spot them, according to a preprint posted on the bioRxiv server in late December.

So far so good. Per the preprint, researches collected 50 real abstracts, 10 each from JAMA, NEJM, BMJ, Lancet, and Nature Medicine, then asked ChatGPT to generate a new abstract out of each article’s title and journal name. They ended up with 100 abstracts, half of them AI-generated, that they were able to analyze using 3 methods: a plagiarism detector, an AI detector [Note: Or, to be more precise, the GPT-2 Output Detector. Note that ChatGPT is based on GPT-3. ] , and blinded [Note: The preferred term nowadays seems to be masked over blinded, but either way you are bound to have funny-slash-distrubing associations pop into your head. ] human reviewers.

You can click through the link to read the outcomes, but per the pre-print’s own conclusion:

The generated abstracts do not alarm plagiarism-detection models, as the text is generated anew, ⊕ [Note: Emphasis mine. ] but can often be detected using AI detection models, and identified by a blinded human reviewer.

So the “can often be detected” from the preprint itself becomes “often unable to be spotted” in the hands of a crafty human reporter. Gotcha.

Of course, no alarmist article is complete without some color comentary:

“I am very worried,” says Sandra Wachter, who studies technology and regulation at the University of Oxford, UK, and was not involved in the research. “If we’re now in a situation where the experts are not able to determine what’s true or not, we lose the middleman that we desperately need to guide us through complicated topics,” she adds.

We have always and forever will be in a situation where everyone — expert or not — had to engage their critical thinking to determine whether data presented are true and important, true but unimportant, true but misinterpreted, fragile, exagerated, overblown, or just plain fake. AI making it easier for the unscrupulous to do what they would have done anyway does not change the equation by an Earth-shattering amount.

Look, some people can’t handle a blank page but are good at editing, even if it means completely replacing the original text. In the olden days of 6 months ago trainees had no other recourse but to grind their teeth and just get on with it, hoping that at some point in their careers they will have trainees of their own writing those pesky first drafts. ChatGPT seems like a godsend for them. Whether what’s sent to journals for publication or posted on a pre-print server is real, fake, nonsense or profound still depends on the person doing the submitting.

Some side observations in no particular order:

• I have no issue with the pre-print itself, which I hope and trust will find a good home.
• Why does Nature deem the work important enough to cover in a news article, but not important enough to publish in one of its own journals?
• For an online news article, it is sadly lacking in that great breakthrough from six decades ago, the hyperlink. Even the URL for the pre-print itself is given as an un-clickable footnote. And no mention of the online and freely accessible plagiarism and AI detection tools.
• Nature’s news department is on a roll.

If ChatGPT and other predicitive AIs kill Google, it won’t be because they are better at search, but because the loads and loads of inane content they generate make search results unusable.

Back to human-curated website lists it is.

Craig Hockenberry:

A lot of folks appreciated the visual design of our Twitter app. And we are proud of that.

We’re equally proud of the things you don’t see.

Which reminds me of why Frasier was so good: for all the jokes they didn’t make.

An update on an update: the projector is en route to… Alabama, I think, leaving us without a living room screen for the next couple of weeks.

It is fun to not require any willpower not to plop down on the couch and watch something, and have every night be a board game night.

The headline: ChatGPT appears to pass medical school exams, educators rethinking assessments.

The article:

• They were mock, abbreviated exams,
• done incorrectly, ⊕ [Note: There are no open-ended questions on the real USMLE. ]
• which it didn’t actually pass,
• and which were reported in a pre-print. ⊕ [Note: Which isn’t a complete knock against the study per se, but even a glance at it shows that some questionable choices have been made regarding the scope — there were only 376 publicly available questions instead of more than a 1,000 on the real exams — and the methods used to ensure the publicly available questions hadn’t already been indexed by the ChatGPT training algorithm. ]

To be clear: this is my complaining about misleading headlines, not saying that predictive AI wouldn’t at some point be able to ace the USMLE, that point not being now, for reasons stated above. And let’s not even get into whether having a high USMLE score means anything other than the person achieving a high score being a good test-taker (it doesn’t).

And with that, the Twitter chapter of my life has closed.

May 2008 to January 2023. Not a bad run, considering.

Tech trouble update: LG’s only way of communicating is via phone calls at unpredictable times from unpredictable numbers without the option to call back. The projector repair is therefore still up in the air.

On the other hand, since I had to turn off silencing unknown callers, I have become exposed to the wasteland that is the American phone system. It is like a George Saunders short story: robots, aliens, and the ocasional lifeless middle-aged salesperson. LG, why do you torture your customers?

Things that the Arc browser does well:

• Mini-browser for opening URLs in other apps
• Built-in shortcut for copying current window’s URL
• Split-screen browsing
• Pinned tabs and Spaces instead of bookmarks

The one thing keeping me from using it full-time:

• Can’t open bookmarklets

We are only 6 months away from the 10-year (!?) anniversary of Vesper, an app that not only still works on iOS 16, but feels more at home there than on the iOS 6 it was made for. Kudos to @brentsimmons, Dave Wiskus and @gruber for seeing the future. Too bad iOS 7 overshot.

A phenomenon so common, it has its own Wikipedia entry:

Link rot (also called link death, link breaking, or reference rot) is the phenomenon of hyperlinks tending over time to cease to point to their originally targeted file, web page, or server due to that resource being relocated to a new address or becoming permanently unavailable.

Sifting through dozens of old blog posts as I transfer them to Micro.blog, a few things are becoming evident. Having a newborn in the household is not conducive to writing. The period from Thanksgiving to New Year’s Day is. And most of the internet I have linked to in the past no longer exists.

Take this short, 8-year-old text about vim. It mentions one podcast and two blogs; none are still around at their original location. The podcast was Technical Difficulties which, if memory serves, was a podcast of Gabe Weatherhead and Erik Hess that ran for 2 years from 2013–2014 before disappearing into oblivion. One of the two blogs was Dr. Bunsen by Seth Brown: also gone, but at least available through the Wayback Machine. The second was from Steve Losh whose website is either down or having temporary difficulties, but in any case unavailable on Wayback.

So this little expedition through just three links took me a good 15 minutes; updating all of the old posts with new links and explanations like this one would not be the best use of anyone’s time. But what are the alternatives?

Gwern Branwen’s website comes to mind, as he goes as far as hosting complete pages on his own server while using icons to point to the original URLs. The afformentioned Wayback Machine also hosts web page snapshots. Would a script converting original URLs to their archived counterpart be hard to find, if not make?

Those are not bad ideas — for a digital garden-type project. For an effemeral blog such as this one, the effort-to-benefit ratio leans the way of my learning to live with link rot. So it goes.