I really want to use AI like llama, ChatGTP, midjourney etc. for something productive. But over the last year the only thing I found use for it was to propose places to go as a family on our Hokaido Japan journey. There were great proposals for places to go.
But perhaps you guys have some great use cases for AI in your life?
I don’t. Played with it a bit but as a capable writer and coder I don’t find it fills a need and just shifts the effort from composition (which I enjoy) to editing and review (which I don’t).
Mostly the same. I tried ChatGPT a few times to get it to generate some code, but mostly it produced code that didn’t even compile and when I asked it to fix it, it created code that didn’t compile in a different way. I enjoy writing code on my own a lot more than having to review some pre-generated code.
Though I use it as a glorified Google sometimes and that is not even so bad.
Not much. I totally agree with Linus Torvalds in that AIs are just overhyped autocorrects on steroids, and I despise that the artwork generators are all based on theft.
Pretty much all I use them for is to make my life easier at work, like turning a quick draft into a formal email.
The LLMs for text are also based on “theft”. They’re just much better at hiding it because they have a multitude more source material. Still, it does sometimes happen that they quote a source article verbatim.
But yeah basically they’re just really good copy/paste engines that work with statistical analysis to determine the most likely answer based on what’s written in basically the whole internet :P It’s a bit hard to explain sometimes to people who think that the AI really “thinks”. I always say: If that were the case, why is the response to a really complicated question just as fast as a simple one? The wait is just based on the length of the output.
In terms of the “theft” I think it’s similar ethically to google cache though.
It’s a bit hard to explain sometimes to people who think that the AI really “thinks”
If I had the patience, I’d try to explain the Chinese Room though experiment to the people that misunderstand AIs. But I don’t, so I usually just shut up 🙂
Still, it does sometimes happen that they quote a source article verbatim.
I’m hoping it’ll quote the license I put in my comments (should my text ever be included in the training set) and gets somebody in trouble. But yeah, transformed anything is difficult undo to see what the source material was, so commercial LLMs can mostly just get away with it.
I totally agree with Linus Torvalds in that AIs are just overhyped autocorrects on steroids
Did he say that? I hope he didn’t mean all kinds of AI. While “overhyped autocorrect on steroids” might be a funny way to describe sequence predictors / generators like transformer models, recurrent neural networks or some reinforcement learning type AIs, it’s not so true for classificators, like the classic feed-forward network (which are part of the building blocks of transformers, btw), or convolutional neural networks, or unsupervised learning methods like clustering algorithms or principal component analysis. Then there are evolutionary algorithms and there are reasoning AIs like bayesan nets and so much much much more different kinds of ML/AI models and algorithms.
It would just show a vast lack of understanding if someone would judge an entire discipline that simply.
There is literally no “artificial intelligence” in any of this. It would show a vast degree of BS, hype, and obfuscation to promote data, statistics, and other computations as “intelligence”.
You are literally wrong. Nice article, don’t see how that’s relevant though.
Could it be, that you don’t know what “intelligence” is? And what falls under definitions of the “artificial” part in “artificial intelligence”? Maybe you do know, but have a different stance on this. It would be good to make those definitions clear before arguing about it further.
From my point of view, the aforementioned branches, are all important parts of the field of artificial intelligence.
I don’t and the energy consumption of public AI services is a stopper for “testing and playing around”. I think I’ll just wait until it takes over the world as advertised.
Nope, nothing. There doesn’t honestly seem to be anything I’d use it for, even then I wouldn’t wanna support it as long as it uses Data its gotten by basically stealing. Maybe once that has gotten better I’ll look more into it, but at the current moment I just don’t have the heart to support it
They take what we make, be it art or Text without our or anyones consent, to me thats stealing something. And yes, there are AI Tools fully build on public Domain and open source things, but those are at the moment, few and far between.
This article by Kit Walsh, a senior staff attorney at the EFF, and this one by Katherine Klosek, the director of information policy and federal relations at the Association of Research Libraries are a good place to start.
Summarising articles / extracting information / transforming it according to my needs. Everyone knows LLM-bssed summaries are great, but not many folks utilise them to their full extent. For instance, yesterday, Sony published a blog piece on how a bunch of games were discounted on the PlayStation store. This was like a really long list that I couldn’t be bothered reading, so I asked ChatGPT to display just the genres that I’m interested in, and sort them according to popularity. Another example is parsing changelogs for software releases, sometimes some of them are really long (and not sorted properly - maybe just a dump of commit messages), so I’d ask it to summarise the changes, maybe only show me new feature additions, or any breaking changes etc.
Translations. I find ChatGPT excellent at translating Asian languages - expecially all the esoteric terms used in badly-translated Chinese webcomics. I feed in the pinyin word and provide context, and ChatGPT tells me what it means in that context, and also provides alternate translations. This is a 100 times better than just using Google Translate or whatever dumb dictionary-based translator, because context is everything in Asian languages.
Oh that reminds me of another use of it last year. I let it translate some official divorce papers from Korean to German and then let a human read through it and give it a stamp of approval. Payed $5 for the stamp instead $70 for the translation.
Naming things in programming is a solved problem now. You can just name it Thingy, and then ask Copilot Chat what it should be called when you’re done implementing it
I’m not a programmer but I work with IT and I regularly need complex shell scripts. ChatGPT has enabled me to skip the largest portion of slamming my head against the wall by writing the script for me and then I can tune it if it gets something wrong. Saves me hours and hours of my life.
Dito, although probably not in the same way you mean :D I’ve actually noticed that I respond stronger to erotic short stories than straight up videos or images, so I use AI for basically erotic fantasy chatting. Some of them can actually generate images to show surroundings or chars during conversations and weave them into the chat.
Currently leeching on CrushOn.ai and occasionally on pephop.ai I think the main issue with a lot of the bots is, that they are way too willing, no matter what their character description or tags say. Chars that actually keep their pants on for more than a couple dozen messages are more enticing to me.
I generally run my self hosted AI. No need my smut to be on corporate servers lol. I think since the stufd I run sre multipurpose models chars tend to focus less on sex if the character description says so
Well that depends. I am no expert by any means and am very much behind on what’s latst and greatest. you will need KoboldAI and maybe SillyTavern too. And you can get models to run on it from huggingface. First how much vram do you have? Before I can (try to) reccomend a spesific model we need to know what you can run.
Ah, I thought only the training was gpu intensive. I currently only have the gpu that comes with the i7 in my server. Guess I’ll just have to continue to use my sites until I can afford to throw a gpu on my server :D
I find them neat, but there’s just too many issues I can’t overlook.
The environmental impact of these technologies is immense, and growing exponentially.
A vast amount of the training data used for the big llms and image generators is not in the public domain, which is at best ethically grey but at worst just blatantly exploiting artists and other professionals.
If there existed some alternatives to the big names that avoided both of these issues, I’d love to use them for code autocomplete and image generation for ttrpgs, but as it stands the moral cost is too high.
I use it to generate code documentation because I’m incapable of documenting things without sounding like a condescending ass. Paste in a function, tell it to produce docstrings and doctests, then edit the hell out of it to sound more human and use actual data in the tests.
Its also great for readmes. I have a template that I follow for that and only work on one section at a time.
I used to spend hours agonizing over documenting things because I couldn't get the tone right, or in over explained, or some other stupid shit.
Now I give my llamafile the code, it gives me a reasonable set of documentation, I edit the documentation because the LLM isn't perfect, and I'm done in 10 minutes.
Over-explaining is my biggest issue. I’m entirely self taught and the trash quality of certain softwares with non-descriptive variable and function names sort of steered me towards clearly naming things (sometimes verbosely). That has the unfortunate side effect of repetition when documenting and it comes across as sarcastic or condescending when proofreading.
Its far easier to have a machine do it than to second-guess every sentence.
You mentioned a llamafile, is that offline? I’m using GPT-4 at the moment because my partner has a subscription. If so, I maaaay have to check it out ^^
Llamafile runs entirely on your machine. The largest one I can run locally is Mistral-7B and Wizardcoder 13B. They seem to be on par with chatgpt-3, but that's okay for my purposes.
Its also great for readmes. I have a template that I follow for that and only work on one section at a time.
Templates in sections are somewhere where it shines. I set up a template for giving information about a song – tempo, scales used and applicable overlapping ones, and other misc stuff. It’s really nice for just wanting to get going, it’s yet to be inaccurate. It’s quite nice, having a fast database that’s mostly accurate. I do scrutinize it, but honestly even if it were to be wrong one day, it’s just music and the scale being “wrong” can only be so wrong anyhow.
One of my favorite things to do is pass my speech into it and have it rewrite with fog index “#”. Really helps with speaking to varied audiences about the same topic.
Mostly for finding information that for whatever reason can be difficult to find using search engines. For example, I've used ChatGPT to ask spoiler-free questions about plot points in books I'm reading, which has worked rather well. It hasn't spoiled me yet, but rather tells me that giving more information would be a spoiler.
Last time I tried to look something up on Google, carefully, I got a massive spoiler for the end of the entire book series.
I also use it for code-related questions at times, but very rarely, and mostly when using a language I'm not used to. Such as when I wrote an expect script for the first (and perhaps only) time recently.
So many times I wanted to know the name of an actor who played a character after the first episode and the top result was something like “[Character Name] (deceased)” or " Villain: [Character Name]."
I've found it useful for TTRPGs too. Art generators are certainly helpful for character portraits, I also find ChatGPT can be useful for lots of other things. I've had pretty mediocre results trying to get it to generate a whole adventure but if you give it tight enough parameters then it can flesh out content for you - ranging from NPC name ideas, to ideas for custom magic items, to whole sections of dialogue.
You can give it a plot hook you have in mind and ask it to generate ideas for a three-act structure and encounter summary to go with it (helpful when brainstorming the party's next adventure), or you can give it an overview of an encounter you have in mind and ask it to flesh out the encounter - GPT4 is reasonably good at a lot of this, I just wouldn't ask it to go the whole way from start to finish in adventure design as it starts to introduce inconsistencies.
You also need to be ready to take what it gives you as a starting point for editing rather than a finished product. For example, if I ask it to come up with scene descriptions in D&D then it has a disproportionate tendency to come up with things that are 'bioluminescent' - little tells like that which show it's AI generated.
Overall - you can use it as a tool for a busy DM that can free you up to focus on the more important aspects of designing your adventure. But you need to remember it's just a tool, don't think you can outsource the whole thing to it and remember it's only as helpful as how you try to use it.
I find a ton of uses for quick Python scripts hammered out with Bing Chat to get random stuff done.
It's also super useful when brainstorming and fleshing out stuff for the tabletop roleplaying games I run. Just bounce ideas off it, have it write monologues, etc.
Add comment