How to Use Chat-GPT
[Trivia: What distinction is shared by Bath, England; Berlin, Germany; and Flagstaff, Arizona? You could arguably include Palermo, Sicily, as well.]
The last year has seen a couple of different Large Language Models—computer programs that learned how to output human language so that it looks like real speech—become publicly accessible. It’s been a remarkable time. I’ve noticed, though, that these models are still being approached by people who aren’t using them optimally.
Think of Chat-GPT like a friend who knows a lot but is also a bit of an overconfident fabulist. Yes, he speaks in perfectly intelligible sentences. Definitely, his perspective can be helpful. Also, sometimes he’s just incredibly misinformed.
This should inform the “research” purposes to which you might be tempted to put Chat-GPT: Chat-GPT is helpful for bringing things to your attention that might be out there. “Hey Chat-GPT, what’s a good book for getting into WWII history?” That could yield some good results. “Has anyone looked into attacking a hurricane with a nuke?” These types of queries are good. This might especially apply when you’re looking for something and have only a vague idea, or the keywords are so vague that they’re all hopelessly un-Google-able. I’ve tried to use it for this purpose many times, sometimes discovering what I can’t remember although often just being served a confidently hallucinated paragraph from a real author that he/she never wrote anywhere.
But you must always resist the temptation to treat Chat-GPT as an oracle for true facts. Chat-GPT should never be taken as a source for discrete information about the world. I’ve discussed repeatedly that even mathematical results, which being a computer it is fully equipped to derive, are not something that it handles with the total accuracy that is expected of a computer. You should not consider this any kind of reference work or database at all.
But again, even if Chat-GPT is not a source of information, it can be a helpful sounding board for information you provide. Handing Chat-GPT a decent amount of text and asking for a précis is a useful endeavor if you’re pressed for time: Chat-GPT is comfortable operating at the level of using words as tokens and rearranging them. The cadence and rhythms that underly human speech are something that it grasps. So, to, can it serve as a proofreader: you can give it text that it could iron out for you. If you can’t find the right way to phrase something or order a sentence, it might well find that answer for you.
There are two areas, though, where treating Chat-GPT as a source of information makes sense. The first is a practical endeavor where you can immediately test the information in a low-stakes and private environment, and if the information is no good then it’s immediately clear that it’s wrong and it’s also no big deal: tech how-to’s. Forget how to do something in Excel? Want to know how to format something in Word? It is so much better to ask Chat-GPT how something is done than trying to Google something with overloaded keywords and being directed to reddit or (shudder) Microsoft’s own forums, where all the top results are from 2012 and address Windows 7 versions of software programs. I usually get the solution from Chat-GPT, and if the solution doesn’t work, no sweat.
The other area is foreign language practice. Since these programs are language models, they understand language. And although learning English alone might seem like a daunting concept, Chat-GPT at least was trained through a breadth of languages. I engaged it in a discussion in Latin recently:
The term “game-changer” is thrown around a lot, but this seems to actually be one. The hardest things about learning a language are carrying a conversation and receiving feedback for your own attempts at participating in the language. This language model offers a way to receive constant, prompt feedback on your own efforts to use it to say things—not in a highly artificial, overly structured way like many computer programs, but in something resembling a conversation. And, of course, it’s readily accessible, so even something like Latin, with its dearth of native speakers,1 is viable.
~
Answer: These are the sites of the initial observations of planets . . . and Pluto, which use to be a planet.
Bath, England: Uranus (1781)
Berlin, Germany: Neptune (September 24, 1846)
Flagstaff, Arizona: Pluto (1930)
(Every planet closer than Uranus has been known since Antiquity.)
Palermo, Sicily is where Ceres was discovered in 1801 (shortly after Uranus). It was considered a planet until other asteroids were discovered, at which point it was reclassified as an asteroid—until 2006, when it was re-reclassified as a member of the new “minor planet” category.
You can thank Vatican City and its rock-bottom birth rates for that problem.