“I rewrote Kafka in COBOL”
Oh man, it’s late here and I thought to myself “How would you rewrite a Kafka novel in COBOL?”… 🥴
(In case, anyone actually isn’t aware, they’re talking of Apache Kafka.)
In general, though, yeah, I also find it cumbersome how much noise these toy projects add. Actually usable software involves so much more than just dumping some code into a repo.
Nevermind that even just useful software requires you to not rewrite existing software in a worse way. You need to actually come up with something novel, which requires tons of design decisions.
Letting the LLM auto-complete those is a lot harder, because 1) you need to actually describe design goals rather than just telling it “do it like Kafka”.
And 2) because those design goals will be wrong every so often, and/or the detail decisions that you outsourced to the LLM. And then you still need to painstakingly find out what those detail decisions were, so that you can correct the decision.
Kind of not surprising with the ruling a year or so ago, where Air Canada had to follow through on what its AI chatbot had told a customer.
Well, and I guess, due to basic logic. Any other webpage has to take responsibility for the content they publish, whether it is written by a human or by an LLM. There’s no good reason why Google should be treated differently here.
Still an interesting development, though. There’s no guaranteed way to make an LLM not say something. I guess, what they could do, is to run a regular script over the output before it’s displayed and then, for example, just not display anything, if those publishers’ names show up in the output.