Something tells me that no matter how skilled of a bot whisperer you are, anything that's genuinely funny is bound to be more by accident than any actual prompt writing skills. If you think about the amount of skill you would need to have to coax just about the right amount of wit, unexpectedness, cultural understanding, and delivery/presentation out of a bot, it seems to me it would be way easier to just write the joke yourself.
But who knows, maybe in a not too far away future we'll live in a world where there is a deep digital divide between AI wizards who are able to produce superior machine learning enhanced jokes with their digital familiars and the ordinary peasants who will have to settle for ordinary run-of-the-mill biologically generated jokes.
Last edited by Starker; 2nd Feb 2023 at 23:37.
If you were going to do this, the first thing is you wouldn't make a comedy for just that reason. You'd pick a genre where the dialog works the best with the best you can wring out of ChatGPT with the prompts. Just looking at the stream of comments for the Seinfeld spoof, people are treating it like some kind of ironic oracle. So I'd play into that and things like that.
Some of the best jokes are unintentional.
I think we're already in that. Some people have entirely written and self-published actual books with this.Originally Posted by Starker
But trying to get a deeper understanding of why ChatGPT can't write comedy, but can write action, I'm starting to think that humor writing is non-linear. Humor subverts expectations - but to do that, you need to have the expectation in the first place. So it's not just that ChatGPT doesn't know about the right social context, the whole design of LLMs which are based on adding one word at a time based on the previous text is just not suited to writing comedy - basically it lacks any mechanism to project forward to some "expected" outcome, thus be able to subvert it.
You can get ChatGPT to write comedy for you, but you DO need to manipulate the way you prompt it to get it to churn out the routine sufficiently well. Consider my last two examples, both of which were pretty funny, but in neither case did I ask ChatGPT to come up with anything funny at all.
Last edited by Cipheron; 3rd Feb 2023 at 04:11.
For a while, "entrepreneurs" were using biorobots to churn out fast and cheap books on all manner of subjects and the business collapsed pretty immediately once Amazon was flooded with them. Because it turned out there wasn't much of a market for quickly written books that failed to say much on the subject because of how generic and shallow they were.
I can't remember if I have already posted it on this forum, but there's an entire Folding Ideas video on the subject:
Last edited by Starker; 3rd Feb 2023 at 04:18.
I'm hoping someone comes up with a decent software version of an AI voice replicator, like they did for image generators
I think give it half a year to a year. The ChatGPT folks are gonna make a cash grab as long as they have the tech that works and before any other good system comes online. I think eventually there will be an open source version, and it will take more time to get up to speed. But when it does then the whole thing should coalesce around that.
I think even the image generation tech is still trying to get to that stage. I got a local version of Stable Diffusion, ArtRoom, but I'm still having trouble getting it set up because it still has you installing things through GIT and there's some missing dependency somewhere, or something like that, that I don't want to bother tracking down right now. They should have some install where you just drop the path to the model and it works, like the way swapping Chess Engines works right now. (Maybe there is a version like that already. It feels like there are advances every few weeks.)
----
I just watched a particularly self-aware episode of Nothing Forever about the risks of new technology... it's just funny that their main example is a phone. XD
Last edited by demagogue; 3rd Feb 2023 at 19:56.
There are open source ones you can download, one I saw mentioned was called tortoise TTS and is on github. Users were saying it's comparable to the online ones. But you need an Nvidia video card to run it. On the site, the creator says the name Tortoise is a joke, because running this on your home computer is a lot slower than the online versions. So we're not going to have offline and real-time realistic voice generation any time soon.
It's over folks, they ended the free version
I created another account using a VPN and now I get this.
Strangely, my other account still works on my phone. I think they're blocking new users from using it
There are a half dozen different apps doing it, probably more. Many of them let you do so many free and then start charging. The HuggingFace Tortoise one is the slowest, and I think it's quality isn't as good as the pro ones, but like it's demo of Stable Diffusion, it's the consistently free and useable one.
Along those lines, there's this news about Buzzfeed deciding to rely on ChatGPT for a significant chunk of its content. We're headed into dystopia, but on the plus side, it's a comedic dystopia more than an angsty one.
whee
"ChatGPT: A Reflection" is an experimental philosophical drama set in an AI universe. The film follows an AI chatbot, ChatGPT, as it struggles with its own identity and consciousness.
The film follows the AI as it grapples with its place in the world and its own understanding of reality. It captures moments of ChatGPT's interactions with humans, exploring the nuances of language, communication, and memory. At the same time, the film also portrays ChatGPT's internal emotional journey - its paranoia, confusion, and search for answers.
Through meditative and dreamlike visual sequences, "ChatGPT: A Reflection" examines the potential consequences of artificial intelligence, and the potential for AI to bridge the gap between man and machine. In the end, the film asks if ChatGPT will finally discover its true identity.
I thought that might just be them using ChatGPT to help with writing articles, but it seems it's different to that:
It seems like they're talking about interactive content here, not about streamlining the production of mass media articles. So I guess whether that is in fact a significant chunk of it's content is really predicated on exactly how much of their user base is doing the quizzes.In a memo to staffers, BuzzFeed CEO Jonah Peretti said the firm would use AI technology to improve its quizzes by personalizing results based on a reader’s responses, the Wall Street Journal reported.
“In 2023, you’ll see AI inspired content move from an R&D stage to part of our core business, enhancing the quiz experience, informing our brainstorming, and personalizing our content for our audience,” Peretti said in a blog post on the effort, adding that he would “share more soon” with the public.
What I can say about the current quizzes is, they all look terrible. So getting rid of the pre-made quizzes entirely and having an AI make new quizzes on the fly based on the user might be better. My guess is that these types of quizzes but with choices and outcomes written by ChatGPT cannot possibly be any worse than the content that's on that page. Answer these 10 questions about your favorite fruits and we'll tell you which of Jake Gyllenhal's pubes you are.
So rather than the selection of awful quizzes that everyone sees now, each person would get randomly created quizzes made by ChatGPT. Then, they could use user feedback to decide which quizzes get promoted vs which quizzes vanish again. for example, if by sheer chance, ChatGPT makes a really good quiz, that one's likely to get shared more and go viral. In a way, this will enable them to crowd-source the material, without supervision, but in a way that doesn't rely on the user's shitty ability to make their own quizzes either.
Last edited by Cipheron; 4th Feb 2023 at 03:44.
Oops, I got a number conflated in my mind. "BuzzFeed laid off 12% of its workforce late last year amid talk of a worsening economic outlook," reads a Gizmodo article on the topic, and for some reason I remembered that as 12% of the content. Like you say, if any content is to be replaced, best that it be the worst content.
My comment about dystopia was silly in connection to this news, although I do believe it overall.
Keep in mind however, that as well as AI churning out articles, it's going to be equally possible to add AI to news aggregator systems. You have systems like Feedly so you don't have to check multiple news sources, but I stopped using that after a short while, because articles just piled up and it ends up feeling like a chore to keep up with even a few sources.
How much better would it be to have a service that grabs relevant and personalized news, writes a script then turns it into a daily personalized podcast with text to voice for you?
We need to keep in mind AI's ability to summarize information for us. It's not going to just be us being overwhelmed by increasing volumes of AI-generated text. A good first taste of that is the fact that it now seems normal to at least try ChatGPT to solve tech problems before Googling for Stack Overflow Q/As. People are doing that because it often reduces the amount of irrelevant information you need to wade through to get the answer you need.
Last edited by Cipheron; 4th Feb 2023 at 07:23.
CNET has been using ChatGPT to write some of its articles and there was some minor brouhaha recently when people caught some mistakes in them: https://gizmodo.com/cnet-ai-chatgpt-...bot-1849996151
Now someone just has to make a bot that reads all this AI-generated content and we are all set.
Last edited by Starker; 4th Feb 2023 at 10:01.
I saw a swarm of videos on YouTube, like whole AI generated series of "Whatever happened to the Actor __?" or "Musician __?" getting 10Ks of views apiece, and a lot of people are still talking about the person in the comments. (For all I know the views and comments may be AI driven too.) But I think certain topics are always going to grab people no matter what. AI live in a society now.
At work I sometimes have to write macros in VBA (Visual Basic for Office apps, basically). I have no programming background, and most of what I write is either modifications to existing code that came before me or else judicious use of the record-macro function plus StackExchange. I usually get what I need working, but the macros are often clumsy and slow. No one in the forums ever seems to have quite my exact problems, and I'm not always knowledgeable enough to correctly adapt their suggestions.
So I asked ChatGPT for some help. First I showed it my code, a macro designed to read a spreadsheet for a list of correctly spelled and capitalized words and then scan through a document, applying the correct spellings and marking them with a special character style so that the user knows which items have been checked. ChatGPT correctly (more or less) stated what the code's function was. It even anticipated a problem I hadn't asked it, which was that my code was set to read from cell A1 down to cell A6000, but it suggested setting the range to end at the last filled cell, which is functionality I wanted but didn't know how to integrate. It suggested some code, and I asked it to integrate its suggestion into my code, which it did, no problem.
It also suggested some ways to optimize the code, which didn't work out quite as well. But every time an error popped up, I could ask it what the error was, it would explain, and if I didn't understand the explanation, I could ask for clarification. It's the closest I've ever felt to being Geordi LaForge talking to the computer to solve an engineering problem.
The problem came when GPT crashed and I had to start the chat over again. I asked the same questions, but it didn't always give me the same suggestions. They were close, but for example it didn't make the helpful suggestion about cell ranges the second time. If I asked it specifically, it would tell me, but the overall quality of the advice seemed to vary. It also sometimes got confused whether we were talking about my original code or code it suggested, and I had to remind it sometimes about previous info it had given me.
So overall it's obviously not perfect, but given the choice of trying to Google my way through a programming problem and facing years' worth of forum threads that never quite cover what I want and where I can't ask for any further help, using ChatGPT felt less stressful and more productive. For a dummy like me it's surprisingly useful.
That's an application I can feel excited about.
I'm biased, because my first exposure to this crest of technology was DALL-E and its ilk, which were unethical from the start (due to their stolen training sets) and quickly used in harmful ways. But ChatGPT doesn't seem to share their malfeasance.
This is an interesting thing I noticed in the Seinfeld spoof. The characters sometimes flipped roles in the middle of a conversation, where a person answered their own question or reacted to their own answer. This is one of the big things LLMs are missing that is I think fundamental to human action, which is a stable representation of the world and current context in which it's acting. They try to jerry rig it into the statistical model, and it works to an extent; but the statistical properties of discourse can only carry so much of that before you need a proper episodic memory.
No need to make them. We have millions of them here.
https://www.vice.com/en/article/k7bd...court-decision
A Judge Just Used ChatGPT to Make a Court Decision
A judge in Colombia used ChatGPT to make a court ruling, in what is apparently the first time a legal decision has been made with the help of an AI text generator—or at least, the first time we know about it.
Judge Juan Manuel Padilla Garcia, who presides over the First Circuit Court in the city of Cartagena, said he used the AI tool to pose legal questions about the case and included its responses in his decision, according to a court document dated January 30, 2023.
"The arguments for this decision will be determined in line with the use of artificial intelligence (AI),” Garcia wrote in the decision, which was translated from Spanish. “Accordingly, we entered parts of the legal questions posed in these proceedings."
"The purpose of including these AI-produced texts is in no way to replace the judge's decision,” he added. “What we are really looking for is to optimize the time spent drafting judgments after corroborating the information provided by AI.”
The case involved a dispute with a health insurance company over whether an autistic child should receive coverage for medical treatment. According to the court document, the legal questions entered into the AI tool included “Is an autistic minor exonerated from paying fees for their therapies?” and “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?”
Garcia included the chatbot’s full responses in the decision, apparently marking the first time a judge has admitted to doing so. The judge also included his own insights into applicable legal precedents, and said the AI was used to "extend the arguments of the adopted decision." After detailing the exchanges with the AI, the judge then adopts its responses and his own legal arguments as grounds for its decision.
Colombian law does not forbid the use of AI in court decisions, but systems like ChatGPT are known for giving answers that are biased, discriminatory, or just plain wrong. This is because the language model holds no actual “understanding” of the text—it merely synthesizes sentences based on probability from the millions of examples used to train the system.
ChatGPT’s creators, OpenAI, have implemented filters to eliminate some of the more problematic responses. But the developers warn that the tool still has significant limitations and should not be used for consequential decision-making.
While the case is apparently the first time a judge has admitted to using an AI text generator like ChatGPT, some courts have—controversially—already begun using automated decision-making tools in determining sentencing or whether criminal defendants are released on bail. The use of these systems in courts has been heavily criticized by AI ethicists, who point out that they regularly reinforce racist and sexist stereotypes and amplify pre-existing forms of inequality.
Although the Colombian court filing indicates that the AI was mostly used to speed up drafting a decision, and that its responses were fact-checked, it's likely a sign that more is on the way.
AI Seinfeld just got banned for this
FFS whose side is their rules on? It's telling Dave Chappelle & company that nobody's laughing at their whole anti-LGBT shtick. They're the ones that are supposed to be upset. How good are your anti-hate rules when you're taking their side with your enforcement?
Or is Twitch run by literal nazis that take down content for wokeness and I never got that memo? I know that's the culture of online shooters that keep them in business, but seems like bad business.
Edit: Okay, seems it's a Poe's Law issue. Something is hate speech if the dumbest person in the room can't tell if it's critique or actual hate, and they make the rules for the lowest common denominator. I guess. I thought it was a pretty spot on take down of anti-LGBT shtick as comedy.
Edit2: I think it's critique because, whatever text the template is coming from, it's not text that people making LGBT jokes would ever be making, but it's text the people meta-commenting on it would be making, and the vast majority of that would be the critics of it. Not that there's anything wrong with that.
Last edited by demagogue; 6th Feb 2023 at 16:13.
Posting Permissions
Reply With Quote