Lot of great discussion of modern worship music, the depth of lyrics vs. what AI can do, practical uses of AI in songwriting, and more. This is a solid video.
Hallucinating artificial intelligence can tank a court case by creating fake case citations that leave the lawyers open to sanctions or the proceeding itself vulnerable to being overturned, a former litigator said.
Last month, a judge handed down a $5,000 penalty on a law firm representing Colombian airline Avianca Inc., which used ChatGPT to write its legal brief, but the AI included fabricated judicial decisions.
A similar case happened in South Africa, and the judge and magistrate overseeing the cases ripped the law firms in their decisions.
“There is potential harm to the reputation of judges and courts whose names are falsely invoked as authors of the bogus opinions and to the reputation of a party attributed with fictional conduct,” the judge presiding over the Avianca case wrote. “It promotes cynicism about the legal profession and the American judicial system.”
When I was in high school, Internet-powered translation services were just starting to come about. Write up a paper in English, drop it in, select French, and bang goes the donkey. Of course anyone that knew French could tell that the paper wasn’t written in French because of all the obvious mishaps. If you know you know.
It’s much the same now. Relying on AI to do your job requires you then to spend equal amount of time checking over all the work AI did because it’s really super confident that it did the job right. Sometimes you have to be a level or two over your skills to catch the mistakes it made.
This is just one of many reasons I don’t use AI tools for coding yet. They aren’t ready. I’m not sure when they will be ready.
You never asked this question. Guaranteed. What if Elvis sung Baby Got Back?
There were eras in which the work of Christian poets was respected and even lauded. But that was then and this is now. While we still value poetry in the form of songs, most of us pay scant attention to reading or writing poetry. There could be any number of explanations for this, though I am inclined to blame the decline of formal verse (i.e. defined forms of poetry) and the rise of free verse (i.e. neglecting rhyme and meter), much of which is enough to cause the best of us to give up on poetry altogether.
Tim Challies, Poetry of Redemption
I have been trying to will myself back into poetry. I used to consume a lot of poetry. Pretty sure I lost it in my tumultuous twenties. The quote above started my Sunday with lament and awareness that it wasn’t just me seeing poetry’s decline.
And then I read this morning that the rising AIs cannot write poetry. Or do basic arithmatic, which is unsurprisingly interlinked with poetry.
So I decided to try a nonce form and asked ChatGPT to produce a poem with a particular number of stanzas and a set number of stanzas per line. Over and over, it would write a few stanzas with the correct number of lines and then veer off towards the end and produce a much longer stanza. Like it lost count.
The danged robot couldn’t count.
AK Krajewska, Robot without rhyme or rhythm
Some very interesting points follow in the article— which you really should read.
Ted Chiang explained that when large language models (LLMs) are trained, they don’t actually assimilate the underlying principles. Instead, they produce the statistically likely next thing.
These LLMs does actually understand, as they cannot. Well, they can understand, but not truly. Because English doesn’t give us multiple words for understand. In Christian circles, we oft separate head and heart knowledge. These AIs have head knowledge but no heart.
More than that, formal verse is an exercise in applying principles you’ve understood. ChatGPT could produce a statistically likely definition of a sestina based on all the examples of sestina definitions it had come across in its training. To produce a sestina, it would have to have assimilated the principles.
There’s one more reason why LLMs can’t write formal verse, and this one is a little more obvious, though still, I think, worth mentioning. LLMs are trained exclusively on written text. They do not have the sound of words in their training, as far as I know.
Formal verse with meter and rhyme relies on the sound of the words. While you can guess what words are statistically likely to rhyme based on their spelling, it’s only saying them out loud that lets you know if you’ve succeeded.
Our language is far more complex than letters combined. Pronunciation is key to writing poetry. Manipulation of pronunciation too.
I wonder what other effects LLMs will have on literature. Might formal verse in English, which has fallen out of favor since the early 20th century, make a comeback as a prestige form, edging out free verse?
And there was the full circle to Tim Challies. Tim noted that the last hundred years have seen a dearth of poetry. But now it may be what separates us from the AI.
Finally, I wonder if, given that LLMs can produce polished but contentless prose corporate speak, will poetry make a comeback as the form for signaling sincerity? Could you imagine getting a notice of layoffs from your very humane VP in the form of sonnet? I’m not sure it would be a better world but it would be interesting.
It begin nearly a year ago when we entered the IBM Watson Mobile App Developer Challenge with a unique concept: a toy that could learn and grow with a child. Winners of the challenge would receive access to IBM Watson — IBM’s powerful cognitive supercomputer - which is one of the elements necessary to create a truly transformational toy.
Yes, David, Teddy has arrived. Just what we need, a dinosaur with access to IBM’s “powerful cognitive supercomputer” in every home teaching our children to submit to their no-longer-extinct overlords. Cool concept, but I will not be buying one.