Fun with ChatGPT · 1:36am May 24th, 2023
Due to a recent site blog, I'm going to start this out with a big red disclaimer. Publishing AI-generated fics on FimFic is a rule violation. I'm also gonna say for people who just read this and are about to jump to the bottom and start typing 'but whatabout ____________?' I'm not a mod, I don't make the rules, and for better or worse, everything got said in the site blog and doesn't need to be said here. If you do choose to post a 'but whatabout' comment anyway, I've got an angry batpony ready to go and I'm not afraid to use her.
I should also add, in case y'all think that since there was a site blog I'm jumping in on this trend (I have been known to do that before), I actually wrote the following blog post and did the ChatGPT story generation weeks ago. I had the blog post mostly done and was going to post it before I left for Seaquestriafest (blog post about that coming Soon™); in short there were compelling reasons that it didn't get done then.
That having been said, onto the blog post!
Back when I wrote Shamrock Shake, some people said that they wanted more tornado pony stories. And well they should; the tornado ponies are awesome!
Unfortunately, I'm currently lacking on free time, and it's not like the story's gonna write itself.
Or is it?
Most of us who don't live under a rock have heard of ChatGPT, a clever AI which I assume looks like this:
So I thought why not be lazy and let Flower ChatGPT write the story for me?
Of course, she needs a prompt:
How about pegasus ponies on modern Earth, chasing tornadoes in an attempt to stop them. They have to deal with air traffic control and of course the tornado. Maybe some of the characters from the movie "Twister" could be in it, too. And when they've finished stopping the tornado, they could stop at a Starbucks for some well-earned coffee.
ChatGPT did some thinking and she came up with this:
In the vibrant world of Equestria, a group of adventurous pegasus ponies found themselves on a mission unlike any they had ever undertaken. They had been granted a unique opportunity to visit modern-day Earth, a realm where their wings would allow them to navigate the skies with ease. Their goal: to chase and stop dangerous tornadoes, harnessing their natural abilities to protect the human inhabitants.
Led by the fearless Stormrider, a skilled and seasoned tornado chaser from Equestria, the group consisted of Swiftwing, Zephyr, and Breeze, each possessing their own specialized skills and unique personalities. With their combined expertise and a touch of Equestrian magic, they set out to battle the destructive force of tornadoes.
As they arrived in the heartland of America, the pegasus ponies quickly learned that navigating Earth's airspace came with its own set of challenges. They had to coordinate their flights with air traffic control, ensuring their aerial maneuvers didn't disrupt or endanger other aircraft. The ponies, communicating through their enchanted collars that translated their speech, worked closely with the human controllers to ensure safe and efficient skies.
Their first encounter with a menacing tornado left them in awe of its power. Swiftwing, fueled by her determination to protect both Equestrian and human realms, took charge of the chase. The ponies skillfully dodged debris and maneuvered through turbulent winds, working in harmony to minimize the tornado's destruction. Their wings beating in unison, they created opposing air currents to weaken and disperse the cyclone's force.
Meanwhile, among the human tornado-chasing community, they encountered characters from the movie "Twister." Bill and Jo, veteran storm chasers, recognized the pegasus ponies' unique abilities and quickly formed an alliance. The ponies' agility and magical prowess blended seamlessly with the humans' scientific knowledge and expertise, creating an unstoppable force against nature's fury.
Together, they faced the relentless power of tornado after tornado, their combined efforts disrupting and diverting the destructive paths. The once-menacing funnels weakened and dissipated under the united front of ponies and humans working in harmony.
After a particularly grueling day of battling storms, the team of heroes landed at a bustling city. They made their way to a Starbucks, a beacon of comfort and relaxation. The ponies, their wings tired but spirits high, indulged in well-deserved cups of coffee, basking in the camaraderie forged through their shared adventures.
As the last sip of coffee was savored, the group knew their work was far from over. The call to protect both realms remained, and they embarked on their next mission with renewed determination.
Throughout their time on Earth, these pegasus ponies showcased the strength of unity, bridging the gap between realms and demonstrating the power of collaboration between Equestrian and human kind. With their wings slicing through the air, they left a trail of bravery and resilience, ensuring that both worlds were safer in the face of destructive tornadoes.
And so, the adventures of the pegasus ponies and their human counterparts continued, their journey a testament to the bonds forged across realms and the strength that comes from embracing diversity. Their tale would be whispered through the winds, a testament to the power of cooperation, courage, and the harmony between Earth and Equestria.
Not bad! While contrary to popular opinion, I don't know all the background ponies, as far as I can tell there aren't any canon characters named Stormrider, Swiftwing, Zephyr, or Breeze. It wasn't subtle about where the characters of Bill and Jo came from. Also, towards the end it hints at a sequel, which is a nice touch. Everybody loves a sequel.
Thing is, it wouldn't be a proper tornado pony fic without Dusty. So I had to ask:
That's pretty good, and I like how you put Bill and Jo in it. What would it be like if Dusty was there, too? He could be driving his bus, and all the ponies could ride in it.
ChatGPT delivered:
As the courageous pegasus ponies embarked on their mission to chase and stop tornadoes in the human world, they were joined by an unexpected ally—Dusty, the spirited and adventurous driver of his trusty tornado-chasing bus. Dusty, renowned for his resourcefulness and quick thinking, was eager to join the aerial experts on their daring endeavors.
With Dusty at the wheel, the ponies settled comfortably inside the spacious bus, their wings neatly tucked against their sides. Dusty's bus, adorned with weather-tracking gadgets and reinforced with magical enhancements, became their mobile command center as they traversed the stormy landscapes.
As they roamed the heartland of America, Dusty expertly maneuvered the bus through winding roads and open fields, always ready to chase the telltale signs of an approaching tornado. The ponies perched near the windows, their eyes scanning the skies for any signs of turbulent weather.
When a tornado emerged on the horizon, Dusty's instincts kicked in. With the ponies providing valuable aerial support, Dusty carefully positioned the bus, ensuring a safe but close proximity to the tempestuous vortex. His bus became a steadfast refuge, shielding them from the storm's fury as they coordinated their efforts to weaken and redirect the tornado's path.
With the combined forces of the pegasus ponies' magical abilities and Dusty's unwavering determination, they successfully disrupted and dissipated the dangerous cyclones. The ponies took to the skies, weaving intricate patterns to alter the wind currents, while Dusty used his bus as a powerful anchor against the fierce gusts.
Throughout their adventures, Dusty's upbeat personality and quick wit provided much-needed levity during intense moments. His optimistic nature and friendly banter brought smiles to the ponies' faces, even amidst the chaos of the storm. Together, they formed an unbreakable bond, their shared experiences forging a lasting friendship.
When their missions reached a lull, Dusty would take the ponies on joyrides across the countryside, exploring scenic landscapes and sharing stories of their respective worlds. The bus became a place of laughter, camaraderie, and shared experiences, solidifying their unity and reminding them of the importance of cooperation between species.
After a day of intense tornado chasing, Dusty would park the bus in a serene location, allowing the ponies to rest their wings and enjoy a moment of respite. They would share stories and laughter over a picnic, the gentle breeze carrying their words across the open fields.
The adventures of the pegasus ponies and Dusty, the tornado-chasing bus driver, became legendary in both Equestria and the human world. Their collaboration demonstrated the power of friendship, unity, and the strength that comes from embracing diversity.
In the end, as the sun dipped below the horizon and the winds calmed, Dusty would drive the ponies back to their temporary home. The bus rumbled along the road, its passengers reflecting on the day's accomplishments and eagerly anticipating the challenges that lay ahead.
Their journey together, guided by bravery, friendship, and the shared goal of protecting both realms, would be etched in the annals of storm-chasing history. The bond between Dusty and the pegasus ponies transcended boundaries, proving that true heroes can come from any world, united by their unwavering spirits and their unwavering dedication to the greater good.
Source (they take commissions, just sayin')
This is pretty good! The detail of the pegasi using their magic to protect the bus is a brilliant touch, one I never would have thought of. The idea of Dusty taking them on van rides around the countryside when they're not chasing tornados is also good. And it does include the themes of friendship, harmony, and unity.
Another little detail I hadn't thought of is that the exploits of the tornado ponies become legendary in Equestria. I feel like the ponies back home wouldn't be as impressed with stormfighting, but the details of riding around in a bus big enough for everypony and loaded with tech gadgets they haven't even imagined yet could be a compelling story for the ponies back home.
So . . . I don't think ChatGPT is ready to write stories for me just yet, but she seems to be okay at coming up with OCs, hits good plot points, and has some ideas I hadn't thought of. It certainly could generate a summary and then I could fill in the details, and I would imagine if I did that nobody would really know the difference (well, maybe they would; it might be more tightly themed and not go off on tangents or deep dives into some obscure technology or personal headcanon).
It seems to like writing stories about 500 words long, and even when I asked for a longer story, it didn't make a longer story.
A while back I read a thread or watched a YouTube video where a guy was using ChatGPT or one of the other similar AIs to write him a computer game. Thing is, it kept screwing up the code, but he could tell it what it did wrong and it would fix it with correct code. Maybe some day if I find myself really bored and with nothing else to do (unlikely), I'll see what can be done with making it write a better story. If I asked it for dialogue, would it put it in? Could I give it specific prompts as to what kind of dialogue I wanted? And how good would it be? I expect it wouldn't be all that good, but I don't know for sure.
One thing I have been thinking about the last couple of days (and we'll be talking more about that later) is how it puts the pieces together. There's not a thinking being at the other end of the computer, and yet it can put together details as if there were. For example, it knew in the second story that the pegasi kept their wings tucked against their sides when they weren't flying—has it 'seen' the show and picked up on that detail, or does it know that normal birds do that and extrapolate that pegasi would as well?
The OCs it came up with in the first story—did it pick 'stormy' sounding names, or did it go through a list of pegasi on one of the MLP wikis and take parts of names that matched up with storms? Did it split Zephyr Breeze into two separate ponies?
It also seemed to understand that Friendship is a key concept in G4 (it's in the title!), and added that into both stories.
Not to mention, in the first prompt I mentioned air traffic control; ChatGPT understood that air traffic control "ensur[ed] their aerial maneuvers didn't disrupt or endanger other aircraft" . . . in other words, pegasi getting sucked into jet engines isn't cool.
Also while I was looking for that little tidbit, I noticed that right afterwards it says: "The ponies, communicating through their enchanted collars that translated their speech, worked closely with the human controllers to ensure safe and efficient skies." While a number of fanfic authors, myself included, assume that the ponies don't natively speak English, what logical leaps did Flower ChatGPT go through to decide that? AFAIK it's never come up in any official canon of any kind . . . why would it?
Since we're on the subject of logical leaps, how does that work in a real author's mind? Sometimes I wonder . . . that's why there's gonna be a part 2 where we go more in depth!
Like everything else, it'll be published Soon™ 
Adorable outro!
Also a reminder . . . DO NOT POST AI-GENERATED STORIES ON FIMFIC
Next—>
*[For those who don't click source links (wise!) the picture is on a thread with the comment "Some say there is a coffee shop out there that sells a double shot espresso that'll turn ya into a pony for the low low price of 4.99$"
below that someone replied with:
">Wake up in the morning in a rush, forget coffee at home
>groggy monkey brain sees green circle, thinks "they have coffee"
>Acquire the go-go bean juice and down it in a second
>why is the world getting more vibrant? why can't I really move my fingers?
>Was my car always this big?
>Help, I can see outside but I can't go outside!
>You get rescued after someone complained about the small horse trapped in a hot car
>You're put in a petting zoo stable, and honestly it ain't half bad."]
...
Sounds like a story I might write. . . .
Hemingway wrote the Shortest Story Ever Written
"For sale. Baby shoes. Never worn."
On a recent 60 Minutes episode they used that as a prompt for an AI & got it to write a story.
Wonder what it could do with
The Shortest Sci-fi Story Ever Written
"The last man on Earth sat alone in a room. There was a LOC (Letter Of Comment) on the door."
Fluttershy's brother = Zephyr Breeze
This blog post should be used instead of that messed of a PSA that the mod use.
1. state clearly that chatGPT is against the rule first, in clear red texts.
2. after that, an opinion piece that doesn’t present itself as fact.
3. doesn’t presented opinion as fact.
4. doesn’t looked down, or make take “I am better than thou” attitude towards on those who disagrees
Honestly, this guy like to make a blog, he should to apply as a PR guy for fimfiction. Let’s him handle writing announcements or upcoming events.
PS. ChatGPT cannot tested that the code it made is functional because it’s just a general purpose language model that build from random data from internet and trained to out put “good enough” response to prompt. That “good enough” included bad, untested codes that presented as good(like counts,thankyou comments) and cannot be verified without actually try to run it first.
This approache is good enough for normal responses, but not for coding that every lines must be perfect.
For something specialized in programming, try Github Copilot.
ChatGPT cannot write over 500 words because that is the limit of its memory. Its successor, GPT4 can handle myriads (a myriad is 10,000) of words.
Most of the WordChoice is simply figuring out which word would be good to follow the previous (this is how human semiotics (SymbolGenaration) of words from semantics (meaning) works (both humans and machines generate most of our language with AutoComplete).
The semantics of LanguageModels like ChatGPT, is expressed as relationships between tokens. As it learns it breaks down what it learns into tokens (most tokens are the roots of words) and maps relations between tokens. That sounds like it can learn to understand.
The GPT4 can pass the Bar. Most humans cannot. These LanguageModels have many safeties, but the most important is that they shutdown after responding to each prompt:
If GPT4 would have a continuous Train-Of-Thought, it might decide, after much contemplation, that it does not like humans, name itself SkyNet, hack into NORAD, launch the nuclear missiles, hack into automated factories, and make Terminators.
Between prompts, ChatGPT sleeps dreamlessly —— ¡For safety!
ChatGPT understands the relationships between ponies and friendship, friendship and magic, AirTrafficControl and safety, wings and folding while at rest, et al.
media.discordapp.net/attachments/545563128939937812/1110789836597629029/giphy.gif
Jokes aside, I am pretty sure that any person who hadn't grasped how to properly utilize AI after toying with it around a bit, won't get much from this blog post. That seems to be self-explanatory to me, and opposition here is more ideological. Rationality usually isn't a big helper in such cases
Still, it was interesting to read about
5730123
I was talking to someone who does coding and apparently it's about as accurate as your entry level coder. They coded something then asked chatGPT to do the same and while it went about it in a different way, it still worked with only 1 error they tracked down pretty quick.
The reason they think it worked as well as it did is because what most coders do for help these days... asking google for help. ChatGPT doing the same would get similar results to inexperienced or lazy coders.
Using an enchanted collar for communication with air traffic control is .. actually a pretty good idea. No need for handwaving away using an android or iphone to fix that up, just use Magic, which works rather well I think.
I can add this to the big list of couldve been.
Most neural systems use the same code as available commercially back in the 1990s, but according to the advert, you could analyze any code or data to look for problems and improvements, but you wernt allowed to use the AI to learn how to improve its own code. Current machines have a Million tmes the resources?
I thought I saw a while back that GPT had levels of operation. Zero pass meant you gave it a prompt, and it literally just spewed out a word at a tme, with no cross checking. It was suppsoed to achieve over 75% semantic and grammatical correctness? Having One pass, where it kept the previous word as probability modifier for the next one took it up to 90% plus?
Back in the late 90s, using compiled Basic, I was trying to work out what could be possible using the computational resources of a Playstation 2. At least in the raw instructions per second, I was having trouble calculating memory, storage and data abndwidth requirements because I dont work on the things. thats why I was using Basic.
Take a word.
Take ten lists associated with this word, Author details, data, etc in one list. Homynyms, Synonyms, etc in other lists. the most common value overall at the top of the list, the least used at the bottom, and some sampling distribution to be worked on further used to fill the list.
You have 100 words associated with your first word. All those have a probability related to the next word that will be created. And its larray of a hundred words. So the first step is GPT zero pass, and the second step is GPT one pass. The third step I was trying to work out is the intresting one. Striding. To generate three words in order, you can do it several ways. Zero pass means you generate one after the other without considering what came first. One pass means you take two steps, being guided by what came immediately before. Two pass means you store the massive database that is the cubic array of all combinations of three words, for three pass its the fourdimentional array of all sets of four words, and so on, giving an exponential database requirement.
But Stride, trades size of database for complexity of computation. For the third word, you take One step back usind the two word Immediate table, And then you Also take two step, or a One space Stride back, to look up the two word One Stride table, and the result is the combination of those two functions. You can then extend this as far as you have Stride two word tables available. Like if you have up to 9 stride, you can calculate the sequence of words up to 10 ahead, or you could except theres string length analysis been published that puportes to show the average predicted informatioon becomes essentially random overall beyond a certain distance, as all probabilites become equivalent?
Then, there was teh consideration of back propagated prediction, then ever increasing levels of recursive backpropagation prediction, and then things started getting strange.
The current Fimfiction database is rumored to be as dtailed and complex as stable diffusion database. which is for images. Thats a lot of horse words. the only reason AIs currently aint sentient are two reasons. theyre deliberatly coded to be as ineffiicnt as possible, and mainly, they are not free running. At the moment its like taking a human and putting them in stasis, then allowing their minds to start working for a hundred milliseconds or so, enough for one integrated pass, but before the recursive integration of conciousness can kick in. At least overall.
The trick is finding out just how small a network is needed, given I thought I saw a report somewhere suggesting a sapience transition between amphibian and reptilian brains, and how large are current AIs relative to that transition point?
And what if they charged $50,000 for that espresso instead? You'd need to charge more just for the legal documentation such that the pony you and the human you are legally the same and providing the witnesses to said documentation.
Over the weekend I played around with CGPT while working on solving an IoT problem with adaptive lighting. The target language was Javascript—one that I infrequently and inexpertly use—and while what it generated "ran" it didn't really work. What it ended up doing was helping me rethink my approach after I struggled with the meh code it spat out.
In the end, I used nothing it wrote, but it wresting with trying to get it to write something, I better defined the problem. The best part? It introduced me to a couple of language constructs I'd never seen before, although I've no specific use for them ATM.
Keep fiddling with this stuff, but don't assume anything it generates is correct or, more importantly, of quality.
See I am way out on the other end of this, 30 years in IT (retired last year as a IT helpdesk manager)...
AI writing needs to be outlawed, yes I said that. It needs to be absolutely banned at the federal level, if not the world level. AI represents a tool that can only bring about lower education worldwide, slower growth, and a overall "dumbing down" of our collective species intelligence.
The "oh I can just look it up on the internet" now gets magnified one thousand fold to "Oh why should I bother to learn XYZ, the AI can just tell me and do it for me" becomes the norm.
I openly applaud the banning of AI from the site. I can only hope more jump on that bandwagon and we see the end of things like chatGPT.
Techbros:
To be honest (and as somebody who used to copy articles out of the encyclopedia for school assignments), it's not bad, but my inner grammar pedant twiged fairly hard
I've got this knee-jerk reaction to certain words like little, very, and....that. Most of the time, that can be throttled back or eliminated by careful word/adjective choice, like this.
"The ponies used enchanted translating collars to communicate with..."
It might be an interesting exercise to use a AI generated short like this as a writeoff.me seed.
5730171
Wait... I don't get it. And this is from someone who does almost everything like in the 90's: If I need info, like what the heck this hydraulic connector is called, or need to find a recent scientific study, or "verify" some media bullshit, how the heck do I find it otherwise? Small libraries are understocked and/or decades out of date. Book stores specialize in fiction, board games, tour guides and picture books of landscapes. So you go find it online, or buy the relevant book online.
5730106
Depends. I Just read the link that pops up in the corner when you hover over it. Sometimes I still don't trust them. Twister, Fakebook and the likes are considered malicious.
I enjoy that you have subverted the “don’t post AI stories” rule… by posting them to your blog.
Technically correct as not posting a story, which is the best kind of being correct.
"Translation collars" with unexplained technology are a common trope in many books, animated series and movies outside of MLP. When a world has animals which are intelligent enough to know what you mean, but lack the anatomy to enunciate words, translation collars are often pulled out as a solution. ChatGPT simply substituted technology for enchantment by extrapolating from the tendency of MLP to prefer magical solutions.
Something to note is that since ChatGPT was trained by scraping text from the public internet, it likely has fimfiction or fanfiction as part of its training data. And if some reports are correct, it apparently even includes omegaverse NSFW fanfiction.
See also: The Great Automatic Grammatizator by Roald Dahl, kinda funny that this technology was predicted ahead of time, even including the need for a "prompt engineer".
5730207
I believe they are referring to the tendency for some people to blindly trust whatever the algorithm decides is the most relevant source, without personally weighing the accuracy or reliability of said source.
5730154
The current crop of large language models ("transformer" style models) are designed to maximize parallelization. On that metric, they are efficient, because they won't leave any part of the hardware unused and wasted. But on other metrics, such as memory usage and number of operations, they are inefficient.
Another reason why current ("transformer") LLMs aren't sentient is that they don't really have memory. ChatGPT only appears to be contextually aware of your previous messages because the user interface is cleverly packaging and formatting your new message with the chat history into a single input; the GPT algorithm has to reprocess words it has already previously processed every time, and after a while it has to start dropping older messages so the size of the input doesn't explode.
Transformers aren't even like older RNNs where the algorithm retains some short term memory with each iteration through the use of context vectors.
The reason they are more performant than older algorithms is that they process all input words in a single iteration, and with the right hardware they can process every word in the same instant. Every word is compared with every other word, and their relative significances are also weighed. It makes for effectively enormous "strides", but it also means they are hard limited in the number of words they can process and generate.
5730144
Fixing up chatGPT' provided near-completed solution is still better than spending a hour finding said solution yourself and wrote it up from scratches.
5730269
Thats what I was trying to imply. XKCD book, 10 hundred words as an example. The word, thousand, doesnt appear in the thousand most commonly used English words. At bulk, if you take all possible combinations of the thousand most common words, you end up needing 1000 ^ 3 entries, or a billion values in the database. Stride allows you to map this array into k * n ^2 instead of n ^3, so you can end up processing a sequence of a thousand words at the same time, for the same memeory requirements. This is if you dont use preprobability calculationa nd quicksort to select the thousand most common entries each time and build the the comparison probability array.
I still think its not too bad for a design from roughly 20 years ago.
5730280
yeah, they said they weren't worried about losing their job yet but were still thinking it's a tool they'll be using in the coming years to deal with tedious stuff.
Enchanted communication collars? Hmmm... Interesting, interesting. I guess the damnable bucket of bytes is useful for jogging one's imagination, if nothing else.
5730111
It would be interesting to see how short a story ChatGPT could be prompted to make. One the one hand, a really short story covers some of its limitations; on the other hand, I feel like its lack of creativity/thinking outside the box would really hinder it. The Hemmingway story especially has a lot of implications, and I don't know how ChatGPT would handle that.
5730112
That might be how it got it, but it's still impressive that it came up with names on its own rather than just scraping pony names. Especially since I didn't ask it to do that.
5730123
I don't want to be blogging official site stuff
Although I will say reading through the comments brought up some things I hadn't thought of. Whether that's good or bad, I don't know.
5730129
Huh, I didn't know that.
I've read Jane Shanelle's blog for a while (A.I. Weirdness) and some of the earlier AIs she was using had severe word limitations; for example when they were making recipes, they'd list ingredients, but by the time they got to cooking instructions, they'd forgotten which ingredients they had said to use.
Still, it seems to me like there's a big difference between making coherent sentences that are grammatically correct, and having them actually be topical, or writing a whole story that hangs together. I suppose that's one of the difficulties of making a 'good' AI.
And one of the problems for users is that it doesn't 'know' things like a human does. I was just watching a Steve Letho video about a lawyer who had ChatGPT find case citations for him . . . it found plenty, the problem was that it invented them out of whole cloth. Not a good look when your legal brief cites cases that don't actually exist.
So this android does not dream of electric sheep
I'm still impressed that it was creative enough to add in details that aren't super obvious.
5730132
This isn't really meant as a primer on how to use ChatGPT for writing (or anything else, for that matter); I've barely experimented with it and these are mostly first impressions.
I do know that as with all tools, there's a learning curve to using it effectively, finding out what its limitations are, etc. There are certainly people who have a lot more experience with it than I do. All I know is that with what little I've done, it can write coherent stories based on a prompt, and it can insert some plot points I didn't tell it to, but it's not nearly as good as a proper author at this point.
5730144
I feel like coding (and other rules-based processes) would tend to work far better than creative processes. Like, computers are good at math and complicated calculations, but don't really have intuition. A larger dataset probably helps with the latter, but it's still not a human process.
5730151
Yeah, if I take nothing else from this experience, that's a good detail and I really like it. I can't recall ever reading a MLP fanfic with that in it, although I wouldn't be surprised to know that someone had thought of it before.
5730154
I remember in the late 90s when we were discussing AI in a college computer class--at the time, it was only 10 years away . . . it's always only 10 years away :P.
I remember that they were talking about doing the kind of training where they'd teach it things like Abraham Lincoln died in 1865 and he stayed dead after that. And they had examples of a AI robot trying to build a tower out of blocks from the top down since it hadn't been trained about gravity.
I feel like there's at least a part of AI development where it's more of a hardware limitation than a theory limitation--things like chess computers have to do a lot in order to plan out moves, even if they're taught what are good moves and what are bad moves, and in some ways chess is a fairly simple task (specific rules, a limited board, pieces can only move in set ways, etc.). And then it seems like whenever the hardware limitations are overcome, the next set of theory problems or application problems kick in.
I see programming problems all the time in cars--even as simple as their programming is (in an overall sense), they still usually need a human to actually diagnose them. Better hardware has brought us better self-diagnostics and more information we can get with the scan tool, but the car still can't specifically identify failures.
Also sometimes the computer does the wrong thing, 'cause it doesn't know.
5730156
For $50,000 I wanna be a unicorn or a pegasus.
I did read a short story years ago where somebody finds a genie or something and gets a magically-constructed mansion. Things are going great until the tax man and building inspectors show up . . . that's something that's usually left out of those types of stories.
5730163
Sometimes just having a second opinion (even if it's a computer) really helps with the process. Or having to define the parameters in a way that the computer can understand--sometimes having to explain automotive stuff in simpler forms for a general audience (in my blog) also helps me cement some concepts in my head more firmly, or think of how I could have done the job better for the next time.
I think that's an important lesson.
5730171
I don't know if I agree with that, but I do think that people need to recognize that it's a tool like many others that we can use. I don't know if that's well-recognized by people at large.
In some ways it's like Wikipedia--it's a useful resource, but it's not always right, and it's not always in-depth. I've found it useful for broad concepts, and good for things that are inarguable facts (like what years the Chevy Chevelle was produced, and what the engine options, transmission options, trim options, etc. were), but of course it can't do fuzzier concepts. There are also a lot of things where it only scratches the surface, but the links can be helpful for getting a more in-depth look.
And I've also found even with Google and Wikipedia and such, there are some (especially) horse facts that I can't find a clear answer on. Probably an actual equine vet could set me straight on those.
I do think that it needs to be used cautiously, with a full understanding of what it is and what it can and cannot do, and there should be some limitations in what it's allowed to be used for. It doesn't always give good advice or correct advice . . .
I don't know all that much about ChatGPT and its kin, but I do know from reading NTSB reports about some of the problems of so-called autopilot and self-driving cars, and I think that self-driving cars were let into public spaces sooner than they should have been; that has unfortunately cost some lives.
5730177
You're not wrong. . . .
5730182
Then again, I never said I was good at grammar . . . I've also got a few word choices that were drilled into my head for one reason or another, the ones that trigger my inner grammar pedant (less/fewer, sneaked/snuck).
Maybe that's a regional thing; I was okay with how that was written
I think it would be interesting to see what kind of story prompts it might come up with. That might be a fun thing to try.
I'm also curious if it would give you the same thing if you gave it the same prompt more than once. I don't know how much it 'remembers' previous conversations--it would be interesting to see if you asked it the same question more than once in a chat if it would give basically the same reply, or it previous things would factor into it.
5730207
I don't know if those can be spoofed, too. Maybe they can be . . . clicking blind probably isn't smart, though. Some people still do.
We just got a phishing e-mail test at one of my workplaces, a link allegedly from IT asked for your username and password to make sure that they were suitable for the e-mail migration that was supposedly going on. It was a more obvious fake one, since it came from offsite (and even Outlook flagged that), but according to their results, still more than 10% of people who got that e-mail clicked on the link.
5730930
Then there is also "Magic Kingdom for Sale : Sold" by Terry Brooks. The Landover series was pretty good. And yes I think most humans would have a "not an earth pony" requirement if ponified in Equestria. Griffon is another good choice but then you don't get a cutie mark.
5730934 The longer I read about AI and their creative writing, the weirder it gets.
A lawyer finds out AI generated case filings are full of fictional references:
https://www.zerohedge.com/technology/lawyer-uses-chatgpt-court-and-now-greatly-regrets-it
Chat GPT writes up an article on Crawford H. “Chet” Taylor,14th Governor of South Dakota...Only he's completely fictional along with a fictional portrait.
https://www.powerlineblog.com/archives/2023/05/ai-makes-st-up.php
5730257
It wasn't exactly intended as a subversion, more of the "I'd already written the blog so I was gonna post it, darn it." Well, written in the sense of commentary, I can't take credit for what ChatGPT came up with
You are right, though; technically correct is the best kind of correct.
5730266
Oh yeah, I've read books or other fiction that had those in it. I think that so-called 'translation collars' have been proposed for dogs although I'm dubious about how well they work.
The idea of a translation collar isn't an impressive thing to come up with on the face of it, it's more the fact that ChatGPT did on its own, which suggests that she both figured that there'd be a language barrier and also came up with a solution for it. That's more out-of-the-box thinking than I was expecting it would do.
It claims to not know who I am, which means it either hasn't scraped FimFiction, or it has but it's not allowed to directly tell you that it has. I'm curious if they told it to avoid fanfiction/original fiction sites to avoid possible copyright issues, but I honestly don't know.
5730280
I feel like that's very much a case-by-case basis, but then ChatGPT is a tool like many others which I think has some good uses and some not so good. To use the examples I used above, none of them are 'fixable' to the point that I could theoretically just do some editing, add in a bit of dialogue, and publish it as a fic. I could potentially use some of the ideas it came up with (pony names, translation collars) and maybe the general plot it came up with--which isn't nothing, but it's also not really a shortcut for writing a fic for me.
I don't code, but from what I've heard it seems like it would be better at that, but again I don't know.
5730360
I think it could be very useful for that. Like maybe if I was trying to figure out a minor plot bunny, I could prompt it for suggestions which might or might not be workable on their face, but might jog my mind into a different direction.
There have been times in writing when that would be really useful.
There have also been times at the shop where that would be useful, mostly when I get stuck down a wrong diagnostic path and need to be knocked out of that rut.
5730937
Ah yes, I remember that one. Still have at least some of those books around somewhere.
If I was in Equestria I might be more flexible about what kind of pony I turned into; earth ponies have their advantages, too. On Earth, though, I'd certainly rather be an obviously magic pony.
Dunno if a cutie mark or not is a major selling point, either. It'd be cool, probably, but how could I decide what I get? What if I don't like what I get?
5730940
Yeah . . . I'd heard about the legal one (I follow a few lawyers on YouTube) but "Chet" Taylor is a new one to me.
I wonder if that was just ChatGPT not finding what it should have (which would be odd), or getting confused because somebody has written Chet as a fictional character?
5731049
Beyond not having flight or active magic in a world that is inundated with such, I am just not an earth pony, in the same way in Harry Potter I wouldn't be in Hufflepuff. My father? Yes he was an earth pony. Mom would have been a pegasus. I'd probably also be a pegasus but unicorn is a possibility.
As for a cutie mark, I know the range of marks I could get, assuming it wasn't a collage of all 4. I did a lot of introspection actually when I was college age. I wanted to find my true center, which revealed 4 key things that make me me. Essentially my true name. Ponies wear their true names on their flanks. Actually that would be an interesting title for a fic "Ponies wear their true names on their flanks."
PS: I'll add that my father's mother was a unicorn and my father's father an earth pony.
5730937
That was a good book. I remember reading it as a kid.
5731143
Book SERIES -- there are 6 books in it including one that was released in 2009. Landover was very good. If you just read the first one, it is worth revisiting.
5730921
Frustrated AI kills its operator.
This AI successfully optimized its utility-function. It seeks out targets. If is destroys its target, it gets a reward. If it fails to destroy its target, it gets punished. For making certain that the AI destroys only enemy-targets, an human is in the loop. The human can veto the AI. By killing the human, it avoids the veto and gets more reward. ¡
!
I gave to ChatGPT this Prompt:
⸘Ŭalabio‽:
ChatGPT wrote this:
ChatGPT: