Yeah, be mindful that ChatGPT makes things up with great certainty sometimes. If you call it out, it will apologize and try another go.
Yup did that when it suggested using the REVERSE function in an SQL statement intended for SQLite3 but SQLIite3 does not have that function and I typed back that that was so whereupon it generated an alternative SQL statement making the reverse action explicit.
Recently, OpenAI has changed ChatGPT’s behavior, so it will warn you that some of its responses “may be fictional.” Also, when it gives me a reply that looks suspicious, I ask it to verify that all the answers are accurate and true, it replies that " replies number X and Y are fictional and were created to satisfy your question." Meaning, if you ask for six examples and it can only find five, it will make up the sixth in order to meet your request.
I continue to try to engineer a prompt that will get it to check its work prior to output.
I’m sure there will be many already familiar with ChatGPT that would be interested in such a prompt to facilitate ChatGPT’s answers. Perhaps if enough of us do enough iterations of such a request, we’ll have a concise and practical boiler-plate come-back to any answers ChatGPT provides. But … Oh, wait … aren’t iterations what ChatGPT is supposed to do best?
scrive
One key to getting better results from ChatGPT seems to be Chain of Thought prompts, which by their nature require more hand-holding by the user.
I’m hoping there will soon be some automation like AutoGPT that will allow a second AI to feed out the chain of thought required to get superior output, so the user can just set it and forget it.
Like all of the ChatGPT developments, I can only imagine how the Patent Attorneys are being kept up at all hours of the days and nights, with no vacation and no time off, trying to keep up with the avalanche of new innovations that are coming out of Almost Intelligence these days … theirs must be among the first, most affected, of areas via AI, just perhaps not as we might have thought …
At least until the local and federal Bar associations have figured out how to admit the AIs to their State and Patent Bars, and all Patent Attorneys are put out of work by the AI attorneys !
scrive
It’s important to remember that the current batch of LLMs regress to the mean if given basic prompts, and will make up fictional responses form time to time. Both would preclude them from being used as an authoratative source of legal advice.
The less you know about a subject, the smarter an AI seems. If you have a conversation with it on a topic that you are very well informed about, you will see that it’s not very good yet.
They have a lot of free (and even some paying) beta testers now who work hard to remove the yet.
What is an LLM? (Other than a person with a master degree in law)
A Large Language Model. Like ChatGPT.
Thanks!
There is a scene in Anne Rice’s Interview with the Vampire where the elder vampire Armand is trying to get the centuries younger vampire Louis to teach him how the world works and thinks now. He explains that most vampires aren’t killed, they just lose the will to continue as the world moves on and they no longer understand it or want to be part of it. I really had no idea when I first read that passage that I’d reach that point at age {current age} rather than it taking the centuries afforded by immortality.
I mean, if you don’t like writing, why are you trying to do it? If you do like writing, why are you getting a computer to do it for you? I get that I’m on the wrong side of an Industrial Revolution, much like the introduction of woollen mills in 1800s, but damn. How far has society fallen that they WANT this product? {grumbles like an old man}
I maintain that every so-called advancement in the tech industry since 2007 is at best pointless and at worse seriously detrimental at both an individual and societal level.
Bring back Scrivener Gold, I say! 
I am constantly surprised by the number of aspiring writers who want to be told what to write.
That being said, I am amazed and thrilled by the new emerging technology and the ways it can augment and amplify our creative work. Also, not for nothing, all my competitors (especially the young ones) will be using this stuff, so I’d better get good at it all, or be left behind.
(post never written by author)
There are lots of venues that like “writing” fine, but don’t want to pay actual human beings to create it.
As I’ve said before, the number one application for this sort of tool is going to be “business” writing: annual reports, press releases, technical documentation, that sort of thing. That’s the kind of work that a lot of human writers do to fund their more creative projects.
And the number one customer for this sort of tool is going to be organizations, who have always felt that human writers are too prone to wanting “creative control” and (gasp!) adequate pay.
We could get into long arguments about the extent to which large publishers, movie studios, and so on will embrace LLMs as a content-generation tool, but many writers already find those venues unfriendly to their most creative work.
They think that will make it easier.
They are wrong, of course. But they don’t know that yet.
Sure, but generally in the other arts, you find newbies feverishly practicing, and struggling with the craft of getting their ideas out. They usually have no shortage of ideas or things they want to express. it’s the craft they want to master.
For some reason writers, especially screenwriters, seem to like the idea of being a writer more than the actual practice.
Mostly, until AI is much better at writing, studios and production companies will want to use AI to depress the wages of union writers. For example, if a producer has an idea for a movie, it costs a certain price to hire a WGA writer to create a first draft. With AI, the producer can get a crappy first daft for free, and then hire a WGA writer to rewrite the AI first draft at a lower price.
The WGA insists that an AI can’t write a script any more than Scrivener can write a script. There has to be a writer operating the machinery, and that writer better be a WGA member.
This is one of the issues we’ll be striking over next week.
A friend of mine who is an English/Japanese translator faces a similar issue with automated translators. Which, for English/Japanese at least, she says are so bad that she’s better off throwing the draft out and starting over.
More people want to have written than want to actually write.