OK, maybe it’s a tad of click-bait 😉 But in working with ChatGPT (still based on GPT-3), I did the equivalent of Googling myself:
I find the mental image of Felix dressed as a Hassidic Jew absolutely hilarious 😂
I did try to search (Google…) for “Accidental Detective”, but it only comes up with a Korean film. Goodreads returns a couple of book results, but nothing involving a Talmudic scholar. My conclusion is that it just sounds plausible, not a real thing.
Interestingly, my wife tried the same thing and got a slightly different result:
Well, at least it got “historical fantasy” and “talented author” right 😜
It’s similar to when I tried generating snippets with ChatGPT. It’s fun, but disappointing given how much stock I put with historical accuracy of Roman life. Back then, Felix had been scratching at the pants under his Toga. (For those who don’t know, unlike a tunic, under a toga you wear the same as under a Scottish kilt ). Now he’s part of books that don’t exist.
I played a bit with using ChatGPT as an editor and it could do some proofreading: it was some effort for me and mostly it changed everything to American English. It also flagged quite a lot of my writings’ content as “inappropriate”, even though it was just a bit of supernatural gore 🙄 I stopped before they suspend my account.
I have no doubt the tools will improve rapidly, as people better than I are working on it. There will be a point soon where it could proofread and maybe copy-edit a work, preserving the elusive “author’s voice”. But even though I use ChatGPT for more than just entertainment and have worked professionally with NLP (Natural Language Processing), Ethical AI, Semantic reasoning, statistical machine learning, etc for several years, for anything that needs a high degree of creativity or accuracy — whether it’s a business document or fiction — I’m still going to rely on a human being to make the final call. It’s the same as the feel of a hand-crafted table vs something from Ikea.
I’m looking forward to playing with the GPT-4 version. While it’s supposedly significantly more accurate and I’m hoping it’ll get things right about me and Felix, I don’t expect the ROI will quite be there in terms of time investment besides dallying and better boilerplate material, mainly because GPT is a model trained on “stuff I found on the interwebs“, so you get what you put in 😉
Still, I have no doubt we’ll see improvements with each generation, though in a Catch-22, you could teach the model the Chicago Manual of Style, but to teach it to work in your own voice you need a good sample of works in your own voice — ones that have been edited correctly… I’ll keep evaluating the technology and apps that come out, both professionally and for writing, but “productivity boost” and ROI will be a calculation for everyone to make in their own context.
Lastly, I strongly recommend you read this NewYorker article for an excellent explanation about the compression of information into the GPT’s large language model, and the inherent limitations in causes.
Anyway, if you’re still reading, until the machines are trained on correct data and evolve actual knowledge, I’ll leave you with this groaner:
If you’d like to read content that’s generated by a human and makes even AIs blush, why not try my free short stories and novels?