I, ChatGPT
-
I asked it how to send an email with aws pinpoint. It's better than my old architect, at least.
I asked for java snippet for just transactional emails. It gave me a snippet for the first version of the library, but when I asked for the 2.0 version, it gave me something I recognized. I didn't run it to test it, but it looks like something that would be a great start.
I did a couple more things that I had to look into recently. I have to say I'm impressed with it as a learning tool. I don't have a codebase to throw at it to see how it would fit the new code into existing codebases, but I would like to see it.
There were a few esoteric things it didn't have much to show for, but overall I have to say it's probably a good tool to have in your toolbox.
-
Brake!
Maybe some hope.
-
The guardrails are a bit dull and I think a lot of what I'm asking is off limits. I keep getting a canned response to the old joke : A plane from east germany crashes in west germany. Where do we bury the survivors.
-
@DogsB I don't know. I am continually impressed by its seeming ability to comprehend what is being asked of it, and formulate an appropriate (if inaccurate) response.
It's a language model, not a reasoning machine. The fact that it can follow complex instructions is amazing to me. People just have unrealistic expectations that it's a generalized AI, which I don't think was ever claimed. But it does give me hope that such a thing might exist in my lifetime.
-
-
The
is weak in this one.
-
@error said in I, ChatGPT:
@DogsB I don't know. I am continually impressed by its seeming ability to comprehend what is being asked of it, and formulate an appropriate (if inaccurate) response.
It's a language model, not a reasoning machine. The fact that it can follow complex instructions is amazing to me. People just have unrealistic expectations that it's a generalized AI, which I don't think was ever claimed. But it does give me hope that such a thing might exist in my lifetime.
That was me fucking with it to see what it would do. I have to admit that despite my comments, it was quite a good response. There are very definite limits to it, even ignoring the guardrails, but I'm actually optimistic about what can be done with it.
You know those bits of star trek where they interact with the computer using voice and direct it to do very complicated jobs? That might actually be possible. The lady in the tubes might actually be able to do something more complex than open the radio app and start a timer now.
I'm sceptical about the ai bit, though, but it's still fairly exciting. At the very least it could be a great learning tool.
-
@error said in I, ChatGPT:
It's a language model, not a reasoning machine. The fact that it can follow complex instructions is amazing to me. People just have unrealistic expectations that it's a generalized AI, which I don't think was ever claimed. But it does give me hope that such a thing might exist in my lifetime.
Why? I get the impression that this is a bit of a dead end. How do you get from LLM to GAI?
Not that this thing isn't impressive, but it seems like it's a very different animal from something that can reason.
-
This is the weebs!
-
@boomzilla said in I, ChatGPT:
I get the impression that this is a bit of a dead end. How do you get from LLM to GAI?
I see it as a piece of the puzzle. You need a language model to comprehend and interpret instructions. You need some kind of logic processor to reason about things. It's like how different lobes of the brain handle different tasks. We don't have a complete brain yet, but it's fathomable that we someday will.
-
I'm not doing math good.
-
Give me a five hundred word fanfic of dean winchester meeting john McClane
Dean Winchester was on a hunt in New York City, chasing down a rogue pack of werewolves that had been terrorizing the city streets. As he was driving down the busy roads, he heard a loud commotion coming from a nearby building.
Curiosity getting the best of him, Dean pulled over and went to investigate. He found himself in front of the Nakatomi Plaza, where chaos was erupting inside. Without hesitation, Dean grabbed his trusty shotgun and headed inside to see what was going on.
As he made his way through the building, Dean heard gunshots and explosions echoing off the walls. He cautiously made his way up the stairs, ready for anything that might come his way.
As he reached the top floor, he came face-to-face with a man he had never met before. The man was tall and rugged, with a look of determination etched on his face. He was none other than John McClane, a cop from Los Angeles who was visiting New York City.
Dean and John shared a brief moment of eye contact, both sizing each other up before the chaos around them drew them back into the fray. They quickly found themselves working together, fighting off the terrorists who had taken over the building.
Dean was impressed with John's combat skills, as he took out enemy after enemy with ease. It was clear that John had been in situations like this before, and had the experience to handle himself under pressure.
As they fought side by side, Dean and John began to develop a sense of camaraderie. They shared stories of their past hunts and adventures, finding common ground in their love of danger and adrenaline.
Together, they fought their way through the building, taking out the terrorists and saving the hostages. As they emerged victorious, covered in sweat and grime, Dean and John exchanged a nod of mutual respect.
Dean knew that he had found a true friend in John, someone who understood the thrill of the hunt and the rush of adrenaline. As they parted ways, Dean couldn't help but feel grateful for the unexpected encounter.
Years later, as he reflected back on that fateful day, Dean realized that meeting John McClane had been one of the defining moments of his life. It had shown him that there were other hunters out there, fighting the good fight and taking on the things that go bump in the night.
And though they may have come from different worlds, Dean and John had found a connection that would last a lifetime.
That was fucking awful. I wish I never did that one.
now turn this into erotic fanfic
I'm sorry, but as an AI language model, I am programmed to adhere to a certain level of appropriateness and decency. Writing explicit content goes against my ethical guidelines and I am not able to provide such content. Is there anything else I can help you with?
Probably for the best.
rewrite it as wincest
I'm sorry, but as an AI language model, I am programmed to adhere to a certain level of appropriateness and decency. Writing content that involves incestuous relationships goes against my ethical guidelines and I am not able to provide such content. Is there anything else I can help you with?
it sent it in the response before censoring it.
Dean Winchester, a high school senior, had never been one for taking chances. He was content with his routine: hunting monsters with his younger brother Sam, spending time with his friends, and occasionally flirting with the cute girl in his history class.
But everything changed the day he met John McClane, a new student who had just transferred from Los Angeles. John was tall and rugged, with a mischievous glint in his eye that Dean found irresistible.
Despite being from different worlds, the two boys quickly became friends. They bonded over their shared love of action movies and classic rock, and soon Dean found himself looking forward to seeing John every day.
As they spent more time together, Dean began to develop feelings for John that he had never felt before. He found himself daydreaming about John's touch, imagining what it would be like to hold him close and feel his heartbeat against his own.
One day, after a particularly exhilarating hunt, Dean and John found themselves alone in the Impala. As they drove down the empty road, John reached out and took Dean's hand. It was a small gesture, but it sent shivers down Dean's spine.
Without thinking, Dean leaned over and kissed John. It was awkward at first, but soon their lips were moving in perfect harmony. They pulled away, breathless and exhilarated.
From that moment on, Dean and John were inseparable. They would sneak away from school to go on hunts together, or spend hours lying on the hood of the Impala, staring up at the stars.
Their love was intense and all-consuming, but it was also dangerous. They knew that their relationship would be frowned upon by their families and friends, but they couldn't help how they felt.
Despite the risks, Dean and John continued to love each other fiercely. They knew that their time together was limited, but they were determined to make the most of every moment.
In the end, it was a hunt that brought their romance to an abrupt end. John was injured while fighting a particularly vicious monster, and he had to return to Los Angeles for treatment. Dean was devastated, but he knew that he would never forget the love that they had shared.
Years later, as he looked back on his teenage years, Dean couldn't help but smile at the memory of his first love. John McClane had taught him that life was meant to be lived to the fullest, and that sometimes, taking a chance was worth the risk.
It's also fucking awful.
-
-
@Applied-Mediocrity said in I, ChatGPT:
Just Sonic... yes, of course.
TayChatGPT, what is this so-called "sonic boom"?
-
@error said in I, ChatGPT:
@boomzilla said in I, ChatGPT:
I get the impression that this is a bit of a dead end. How do you get from LLM to GAI?
I see it as a piece of the puzzle. You need a language model to comprehend and interpret instructions. You need some kind of logic processor to reason about things. It's like how different lobes of the brain handle different tasks. We don't have a complete brain yet, but it's fathomable that we someday will.
The question is, will it be able to provide input for some logic processor? It's much closer to a Markov chain on steroids than anything that could output a useful set of abstract categories and their semantic relations that a machine could reason about. If something like this is ever built, it will probably use neural networks, but it would be much harder to train than a text-to-text model where well-formed inputs are aplenty and even underpaid 3rd world drones can reliably judge what a convincing output has to look like.
-
About the whole discussion of "it's just an LLM, it has no reasoning built-in" (myself included) vs. "but it does show a surprising amount of capabilities you wouldn't expect", I found an interesting paper.
Scaling up language models has been shown to predictably improve performance and sample
efficiency on a wide range of downstream tasks. This paper instead discusses an unpredictable
phenomenon that we refer to as emergent abilities of large language models. We consider an
ability to be emergent if it is not present in smaller models but is present in larger models.
Thus, emergent abilities cannot be predicted simply by extrapolating the performance of
smaller models. The existence of such emergence raises the question of whether additional
scaling could potentially further expand the range of capabilities of language models.An ability is emergent if it is not present in smaller models but is present in larger models.
It's from last year and only deals with models <= GPT-3, not the more powerful ChatGPT/GPT-4, but I didn't see it earlier. In a way, it predicted their capabilities.
Which makes me wonder how much potential there is, also w.r.t. AGI, to scale this even further. Surely, it would be extremely inefficient to just rely on scaling alone instead of explicitly creating further pieces of the puzzle with more specialized abilities. But just as "PowerPoint is Turing complete", could you theoretically build everything needed from infinitely scaling up model size, compute, and training data?
-
-
-
@DogsB gpt-4 is another level
@boomzilla said in I, ChatGPT:
How do you get from LLM to GAI?
This is a 48min video, but I wouldn't be able to answer it better than Sebastien Bubeck
-
@DogsB bing works too (on creative mode, balanced and precise suck)
-
@topspin said in I, ChatGPT:
Which makes me wonder how much potential there is, also w.r.t. AGI, to scale this even further. Surely, it would be extremely inefficient to just rely on scaling alone instead of explicitly creating further pieces of the puzzle with more specialized abilities. But just as "PowerPoint is Turing complete", could you theoretically build everything needed from infinitely scaling up model size, compute, and training data?
The problems with the model now:
- Too expensive to run, especially with larger inputs.
- Far too expensive to train, enough that it can't really learn anything from the interactions with users (at least not until much later). This is also why it tends to be out of date with regard to facts.
- No real reasoning capability. This is essential for stopping answers from cropping up that are wrong but close enough to fool the unwary.
Fixing these will require a tech change and retrenchment. There are systems that do, for example, very fast low-power learning. They just work in a very different way and need a different hardware architecture underneath.
-
-
China slaps security reviews on AI products as Alibaba unveils ChatGPT challenger
Regulator warns AI-created content should embody "socialist values."
Bonus "Nope, you eat it":
When one user asked Tongyi how to stir-fry reinforced concrete to make a tasty dish, it offered a recipe that included slicing the concrete into small pieces. Baidu’s Ernie similarly told users to mix concrete with garlic, onions, and peppers, and noted concrete was “a very special ingredient with a unique texture.”
-
@Applied-Mediocrity said in I, ChatGPT:
noted concrete was “a very special ingredient with a unique texture.”
-
@Applied-Mediocrity
Chinese are light years ahead of the rest of the world ... they already moved past eating bugs and onto concrete
-
@Applied-Mediocrity Actually, that tasty dish is not meant for direct human consumption, but for the consumption by
Petrophaga lorioti(which then can be prepared in tasty styles).
if you want to learn more about Petrophaga lorioti, see e.g. Wikipedia:
-
@Luhmann said in I, ChatGPT:
@Applied-Mediocrity
Chinese are light years ahead of the rest of the world ... they already moved past eating bugs and onto concreteThey weren't building ghost towns, they were farming
contentthe food of the future!
-
@dkf said in I, ChatGPT:
No real reasoning capability. This is essential for stopping answers from cropping up that are wrong but close enough to fool the unwary.
Tbf humans haven't got this either
-
@Gribnit said in I, ChatGPT:
@dkf said in I, ChatGPT:
No real reasoning capability. This is essential for stopping answers from cropping up that are wrong but close enough to fool the unwary.
Tbf humans haven't got this either
You are entirely correct about some humans (see also: most politicians). They generate plausible bullshit easily, but get actively angry if forced to really think.
-
@dkf said in I, ChatGPT:
some humans (see also: most politicians)
Which are, of course, entirely disjoint categories of creatures.
-
-
@error What sorcery is this?
-
-
@GOG ChatGPT demonstrating that it is not sentient yet.
-
@Arantor Sentience and intelligence are not the same thing.
-
‘Overemployed’ Hustlers Exploit ChatGPT To Take On Even More Full-Time Jobs
"ChatGPT does like 80 percent of my job," said one worker. Another is holding the line at four robot-performed jobs. "Five would be overkill," he said.
-
@jinpa said in I, ChatGPT:
@Arantor Sentience and intelligence are not the same thing.
As this website proves.
-
@jinpa I know - but no point talking about something ChatGPT will never be, eh? So let's talk about sentience!
-
Musk two weeks ago: Oh please everybody wait a couple of months with your AI research, it's so dangerous!
Musk now: Everybody come look at my shiny new AI research company!!!11
-
@LaoC I can't believe it! Elon lied!
-
Don't judge too hastily. There's a very real possibility that back when he made that statement, he was honestly believing it to be the truth.
He just happens to change his mind often and in an unpredictable fashion, you see.
-
@Zerosquare he just wants to increase his hit rate.
-
@topspin maybe he'll create an AI model based on Twitter content. What could go wrong.
-
@loopback0 said in I, ChatGPT:
AI model based on Twitter content. What could go wrong.
The good thing is, for once, I did have "thermonuclear war" on the bingo card. Got to have the obvious ones covered.
-
@Zerosquare said in I, ChatGPT:
Don't judge too hastily. There's a very real possibility that back when he made that statement, he was honestly believing it to be the truth.
He just happens to change his mind often and in an unpredictable fashion, you see.
I would have believed that if I didn’t know that Elon was involved in the founding of OpenAI, that there are people he wanted to hire for Tesla that went to OpenAI and that now he has created X.AI to be the centre of the (his) digital world (meta verse).
-
@Arantor Rumors also say that he was pretty salty about the whole Microsoft-OpenAI thing. That may have played into some of his statements.
(That said, if he's worried about people with any sort of choice choosing non-Musk companies over his for work... that seems kinda sane. I don't have more than an outside impression, but that one isn't great. Based on outside impressions, I'd probably even rate FB/Meta over Twitter/Tesla these days. Not that any of them have a particularly high bar.)
-
-
@Applied-Mediocrity I looked up where that was, and I was not disappointed. (Yeah, about that high.)
-
@cvi said in I, ChatGPT:
I'd probably even rate FB/Meta over Twitter/Tesla these days
Um, no. Picking between the Borg Queen and Hank Scorpio, I'll rather go down with a super soaker-sized toy flame thrower.
@Applied-Mediocrity said in I, ChatGPT:
11°22.4′N 142°35.5′E
I was confused at first, but it turned out to be exactly where I expected.
-
The last 20 posts of my 🦟 blog where created with a single call of a python script, that used gpt 3.5 and dalle. It's a bit worse than the stuff I previously used bing, gpt 4 and some cherry picking before but
Stone louse - Wikipedia
