@xiaoyaoyou@pennstac Thanks again guys… so the distillery was Hard Truth and is located in Brown County, IN (or little Nashville).
I highly recommend trying them out but no idea where you all are located…and that’s if you dig on whiskey too of course. I’ll just drop this little nugget here:
I picked up this barrel select of their Sweet Mash Rye, which is my absolute favorite from them. They happened to have this bottle in a Malted Rye special edition. I saw the sig on it and asked about it. Apparently the Master Distiller (I want that title) randomly signs his bottles and sprinkles them on the shelves at no extra cost. Pretty cool and now a collector item that I’ll 2nd guess the right time to pop the cork ha!
Per Google Translate: “It’s important to know that the pain itself is important, and it’s going to be followed by the education system.” As the land grows, I do not mourn nor have any heart.
It’s actually appropriate here, as the fooled buyers will suffer pain and be educated.
My experience is from the 1960’s Letraset days, but I always knew there was more to the story. When in college I took some Spanish linguistic history, I became at least somewhat familiar with 5th century common (vulgar) Latin. (Not the way we use “vulgar” now). This is apparently earlier, derived from Cicero in the 1st century BC (or BCE if you are even slightly woke).
The reason people don’t really think it’s “Latin” is due to its age. @SenyorC do you think you want to move this part of the Lorem Ipsum discussion to the Off Topic?
B.C. is generally used as “Before Christ”, while B.C.E. is “Before Common Era”
I only use the word “woke” to get a knee jerk response. A use that seems rather popular these days.
In my little prospective software-engineer mentoring group … today we talked about generative AI (including true AGI, and LLMs like ChatGPT 3, 3.5 and 4 …) …
The pre-work was to come up with four prompts, on specific themes/tasks, and we’d go over them, make them more useful, specific and accurate. Essentially an interactive introduction to “Prompt Engineering” or “Prompt Construction”. Along with this, they were to come prepared with a couple of questions on how such “AI” might change the workplace.
Following this was the “sad” realization that, based on their prompts and questions, of the 16 people in the current group, ChatGPT 3.5 is already capable of replacing half of them in the hands of someone with relatively basic, but well structured, “Prompt Engineering” abilities.
Which is not to overhype things; context and task matters, as does how you intend to utilize/apply the results … I can trivially set even v.4 tasks that I can address in three lines of very basic code that no amount of prompting/revision yields a correct result to. But for writing papers, analysis and most non-truly-new-acts-of-creation … this thing is alreadythere.
In Neal Stephenson’s novel Fall AI produced drivel designed to keep eyes reading a screen has incapacitated a significant percentage of the population.
As the child in the back seat asks, “Are we there yet?”
Sadly, a bag of Cheetos is more capable than many 1st tier customer service staff. They are hired at minimal rates, have no domain knowledge, and closely follow a script. Even doctors enter symptoms into expert systems – I read 5-10 years ago that those systems were more accurate than the instincts of doctors in making diagnoses. And, IBM’s Deep Blue took the chess crown long ago.
To me, which dystopian SciFi plot do you expect in the present/near future?
Humans kept in a zoo? Vonnegut Slaughterhouse Five
Humans kept as a food supply? Twilight Zone → Simpsons parody
Humans as batteries in a dream world? The Matrix
Humans exterminated by a quack leader? Oh, that was real. Stalin. Hitler. Mao.
My experience with both 3.5 and 4 has been it’s really good at writing documents, but you have to check EVERYTHING it generates, I’ve used to to generate templates for a number of real world tasks, that I’ve then refined.
It’s ability to write software, or even answer uncommon questions is severely limited, and often results in it just making things up. I asked it how to configure the query timeout in redis graph, it responded with an answer that looked like it might be correct, but was completely invented.
I’ve spent time asking it to generate functions for various things, ask it to generate none recursive code to walk a binary tree, and it’ll generate what I’d consider a textbook answer. Ask it to solve a problem that involves one or more libraries, and I’ve seen everything from correct code, to code without error handling, to code that doesn’t compile, code with obvious errors, to code with very subtle errors.
The last of which is obviously the worst possible answer, and of course they all come with very confident responses and it has no idea what it doesn’t know.
It’s better with languages like Javascript or Python than it is with Rust or C, which isn’t really surprising given the training mechanism. It’s better than I would have predicted any software would be at the task in my lifetime, so it’ll be interesting to see how it evolves.
For engineering specifically I do question the ability to move past the quality of the training data, and the lack of some measure of confidence in the answer it’s delivering.
And I’m concerned what happens when it’s in broad use by people who don’t understand the limitations of the software, and just take the answers at face value.
The key to getting accurate data is to create a persona that ChatGPT uses to determine how it responds, what it can consider, tone, style, and so on. This all done through the prompt interface.
You can tell it to exclude data for which there are no established facts, limit it to peer-reviewed sources or academic materials, and even set it up so it will flat out tell you that it cannot answer what you’re asking within the constraints you define on the persona.
Even those tricks are by no means perfect, but they do dramatically reduce how much BS or nonsense it spits out.
For subjects I know well (or well enough) to spot errors or things being made-up (which it will happily synthesize to reach a point, even if the point is wrong), it’s really useful … especially when generating papers or long explanations/explorations. Just requires reading through the output, and some tweaking.
I don’t trust it enough for subjects I don’t know well if I’m trying to support a specific position.
And, as you say, on code, it’ll often hit the mark with well-known problems, and can even do some variation around them, but for something new … it fails hard.
As with most things … it’s a tool … and properly applied it can be a game-changer … poorly applied (or misunderstood) and it can mess things up spectacularly.
Thanks @Torq@generic and @Polygonhell . Y’all inspired me to go for paid subscription and to use version 4. I’m working on a prompt to get it to generate “Ross and Me” stories for FaceBook posts on our business page. Jerry Pournelle, the sci-fi writer describes the Joe and Me format where there is a conversation about something between Me and Joe, Joe being a friend, an expert fisherman, or perhaps a bug-eyed monster. Just a handy vehicle for presenting a story. Ross is my new office administrator.
We’ve got a lot of constraints in posting on Facebook, so I asked the AI to help me generate a suitable prompt that includes enough background information to provide continuity. We’ll see where this goes. I’m lazy about posting, but less so about editing.
Neal Stephenson is one of my favorite writers His stuff is so perceptive and intelligent, plus it’s almost always a lot of fun. The Baroque Cycle was a bit of work sometimes, but still very worthwhile.
I remember reading Snow Crash when it came out, it blew me away! It made my head spin AND it was so much fun, I couldn’t put it down.
Another author I think you’d like if you like Neal Stephenson is Kim Stanley Robinson. I imagine you know of him, his stuff is also amazing, really worth reading.
The best “science fiction” combines great storytelling and character development, and adds a factual science basis for its world building. But I gas on here…
Great to find another Stephenson fan. I too read Snow Crash when it came out, and it started me reading “cyberpunk” - William Gibson, Cory Doctorow, Rudy Rucker. And of course I’ve read Kim Stanley Robinson.
This new novel of Stephenson’s, Fall is a bit of a heavy go in parts, but other parts flow well. I recall reading Cryptonomicon in just under three days in the summer when it came out. Reamde was great, Anathem was a sort of a bizarre page turner. Tip of the hat also to Seveneves and Quicksilver and the Diamond Age. I think Anathem was one of my very favorites, but hey, I come from an academic family.
if you are even slightly woke 
His stuff is so perceptive and intelligent, plus it’s almost always a lot of fun. The Baroque Cycle was a bit of work sometimes, but still very worthwhile.