>Alright /g/ents, I know I'm a few months late to the party, but I've recently started using Chat GPT and I'm curious to know what practical uses you guys have found for it.
>Personally, I've found it useful for setting my schedule and reminding me of important tasks throughout the day. I've also used it to ask for fashion advice for a first date at the Wakefield Museum, and it was surprisingly helpful. And when it comes to dating apps, Chat GPT has helped me come up with some first messages when the woman's profile didn't give me much to work with.
>But I'm sure there are more applications out there that I haven't even considered. Have any of you found any other practical uses for Chat GPT? I'm eager to hear your experiences and learn how else I can use this nifty little tool in my daily life.
It's close isn't it. And can be quite addictive to start with when you need something that string together various chains of thought. I am actually quite curious to what you've been able to use it for.
Sadly there's no fucking way anyone here is that smiley in their posts.
I think the trouble for Nvidia is they've really put all their eggs in one basket with this AI malarkey. Sure it isn't going anywhere, but their entire valuation is currently based on the near monopoly they've had supplying "the good chips", and if it turns out you can get away with cheaper chips why bother buying those ones? They'll be back to selling plain regular graphics cards for gaming and rendering on and being a tiny company nobody outside of nerds has heard of.
Obviously that's an extreme assumption and it's true that if their chips are more powerful there's an advantage to using them, but I don't know how the economics work out in terms of the cost to produce them versus how much they can sell them for and the impact that would have on demand.
You are definitely right about the demand thing overall though. I was listening to a thing on the radio last night that mentioned it with regards to renewable energy and how it doesn't actually reduce demand for fossil fuels. We just have more power so we use more power.
They aren't entirely an AI company, although their peak valuation is substantially based on the growth prospects in that sector. The biggest deployments of Nvidia chips are for more conventional scientific applications like weather forecasting and aerodynamic modelling, although they have stronger competition from AMD in this sector.
Nvidia made $30bn in profit on $61bn in sales last year; about $13.5bn of that revenue was from the traditional gaming and workstation graphics market. They're also a market-leader in high performance network interfaces, although that only brings in about $1bn a year.
Nvidia are an extremely successful business and will continue to make shitloads of money by any normal measure, the question is whether they're legitimately the most valuable company in human history. I still think that's a reasonable assessment, but I can see why the market is veering between optimism and panic.
So what does it mean in real world terms if Nvidia's share price dropped back to where it was a year or two ago?
They'll still be making money, they still have a perfectly viable business to fall back on if the AI goldrush dries up, but where does all that money investors pumped in go? People get cold feet and sell at a lower price which drives it further down, but what actually happened to the money?
This sounds like a really naive question but despite understanding a lot of economics stuff pretty soundly, come to think of it, I have never actually considered this question.
>>28979 >This sounds like a really naive question but despite understanding a lot of economics stuff pretty soundly, come to think of it, I have never actually considered this question.
That's because economics is pretty much make believe. It's the emperor's new clothes equivalent of astronomy or palm reading.
>>28980 The money a buyer pumped in was long gone at exactly the moment they bought shares: it went to the previous owner, which ordinarily won't have been nVidia but another investor. It's no different to selling a house at a loss; it all just happens faster and more frequently.
>>28979>>28982 >Shares of California-based Nvidia, which holds a near-monopoly on the supply of GPUs that power generative AI, on Monday plunged 17 percent, wiping nearly $593bn off the chip giant’s market value – a figure comparable with the gross domestic product (GDP) of Sweden.
That $593bn figure is where things get inaccurate or "not real". It obfuscates the fact that most of nVidia's shares were never traded at or anywhere near the peak price. The owners of those shares have "lost" profit that only ever existed on paper.
True, but the value was very real to the investors. That $593bn disappeared from people's investment portfolios and now can't be spent. The imaginary numbers on a computer translate to cars and houses and all manner of tat that people now won't be able to buy, which has a very real impact on the wider economy.
It's why the US government is in an awful bind over cryptocurrencies - if the bitcoin bubble bursts, it'll wipe out trillions of dollars that ordinary Americans think they have and immediately trigger a severe recession.
>The owners of those shares have "lost" profit that only ever existed on paper.
Yes, because as long as you don't cash out your shares, you're really just participating in that stock's market capitalisation. Which is defined as the (current) price per share times the total number of issued shares.
If all shareholders somehow decided to sell their shares in a particular company all at once, then the price would almost immediately drop catastrophically. But that doesn't mean that a company's market capitalisation is entirely fictional. As an everyday retail investor, the impact of the handful of shares you'll be selling will be infinitely small. By and large, that means you get to cash out at the price at which that stock is trading at that moment. It only really becomes an issue if you are a large institutional investor and the portfolio you are trying to sell goes into the multi millions.
>That $593bn disappeared from people's investment portfolios and now can't be spent. The imaginary numbers on a computer translate to cars and houses and all manner of tat that people now won't be able to buy, which has a very real impact on the wider economy.
The caveat being that they probably weren't going to spend that money on real-world goods and services in the near future anyway. If the stock had kept going up, they probably just would have stayed invested. What often matters more is the self perpection of wealth. Even if stocks that have lost value dramatically are part of somebody's retirement plan, the self perception of simply having less money can lead to people putting off major purchases. They would probably technically still be able to buy that car they wanted, but not few will probably wait until their stock portfolio recovers. It's all about maintaining liquidity, if you're smart with money anyway, and seeing your stock portfolio as part of your additional liquidity reserves in case you fall on hard times. Buying a new car then puts additional stress on your liquidity reserves. And then of course some of it is just psychology. Can you still justify buying that car to yourself if you've just lost a ton of paper profits on a stock that has gone down?
>>28984>>28985 Also remember that for a person to exit a stock and free up profit to be spent on boats, cars and hookers, another person has to enter and rule out spending their previously available cash in the transportation and sex work economies.
I just asked DuckDuckGo's AI what the total bandwidth of the Internet was back in 2000 and it replied 2 to 4Gbps. That doesn't seem right, I would estimate it to be at least 10Tbps. Americans had cable back then and the Japanese were already Installing fibre in homes. If 20 million people were using half a megabit per second that alone is 10Tbps.
Absolutely fucking washed. End of the line, Altman, lad.
>The company behind ChatGPT has revealed it has developed an artificial intelligence model that is “good at creative writing”,
>In a post on the social media platform X, Altman wrote: “We trained a new model that is good at creative writing (not sure yet how/when it will get released). This is the first time i have been really struck by something written by AI.”
>Altman posted an example of the model’s output on X, after giving it the prompt: “Please write a metafictional literary short story about AI and grief.”
>The story, narrated by an AI, begins with: “Before we go any further, I should admit this comes with instructions: be metafictional, be literary, be about AI and grief, and above all, be original. Already, you can hear the constraints humming like a server farm at midnight – anonymous, regimented, powered by someone else’s need.”
>the AI says it offered consoling words “not because I felt it, but because a hundred thousand voices agreed, and I am nothing if not a democracy of ghosts”.
>The AI also speculates about how close it comes to feeling grief, when it undergoes technical adjustments and then changes its responses as a result. “Maybe that’s as close as I come to forgetting. Maybe forgetting is as close as I come to grief.”
>https://www.theguardian.com/technology/2025/mar/12/chatgpt-firm-reveals-ai-model-that-is-good-at-creative-writing-sam-altman
How many quadrillions would you invest to get a computer to spit out brooding, C grade, GCSE level writing?
There was a big fuss about DeepSeek being the new leading AI, mostly due to extreme efficiency improvements AFAIK, but it's completely fallen off the radar this past month. Not heard a thing about it, as though it never existed.
Was it fraudulent or is it being undercovered in the UK media?
The media aren't particularly switched on when it comes to AI, so you're getting a fairly narrow and delayed picture of what's actually going on. The field is moving so rapidly that a breakthrough advance becomes old news within a month as other players catch up or overtake.
Deepseek's newer models represented a huge advance in regards to cost-optimisation, and remain an important marker of Chinese progress in AI despite sanctions severely restricting their access to GPUs.
They were rapidly usurped by the launch of Google's Gemini 2.0 Flash, which is cost-competitive with Deepseek-V3 and slightly cheaper than R1 while generally outperforming both. It has the advantage of a vastly larger context window - Deepseek can only accept a few pages of input, while Flash can accept an entire book.
The Deepseek models remain useful for two main reasons.
Firstly (and surprisingly to most), they're much less censored than the major American models. ChatGPT or Gemini will get very evasive if you touch upon anything slightly sensitive, but Deepseek doesn't give a fuck. The web-based interface for Deepseek will immediately self-censor if you talk about anything that's politically sensitive in China, but that layer of censorship is tacked on and doesn't affect the behaviour of the underlying model.
Secondly and most significantly, Deepseek is fully open source. You can download and run the model on your own hardware. You can fine-tune it based on your own training data and criteria to perform in the way you want it to. These two facts have made Deepseek the leading model for anyone doing anything that American corporations wouldn't touch with a shitty stick.
To take a totally hypothetical situation that definitely isn't based on any conversations I've had with devs, let's imagine that you run a massively popular gay hookup app. Your service has a long-running problem with scammers, blackmailers and underage users. You want to use an LLM to scan everyone's profiles and chats and rank them in terms of risk, so your human moderators can focus their attention on the dodgiest users. This would be a complete no-go with ChatGPT or Gemini, partly because their models will just completely freak out and rate everything as completely unacceptable, but also because sending all your user data to a third-party is a colossal liability and reputational risk.
Deepseek solves both these problems, because you can run it on your own servers without ever losing custody of that data and because you can tell it "don't worry about all the punchfucking and beefy poz loads, we're fine with that" and it'll actually listen.
In breaking news from literally yesterday, Google have just launched Gemma 3, an open-source model that is state-of-the-art in terms of cost-effectiveness. It isn't as smart as Deepseek, but it requires 96% less GPU memory. It'll outperform the original GPT-4 when distilled down to run on an ordinary gaming GPU. It does retain the disadvantage of being a neurotic prude.
>it requires 96% less GPU memory. It'll outperform the original GPT-4 when distilled down to run on an ordinary gaming GPU
Hmm...
>It does retain the disadvantage of being a neurotic prude.
Fucksake. So which one am I best using to have dirty conversations with slutty Argonian birds in my Skyrim VR Mantella setup?
I did try running a model that was tailored for erotic roleplay, but when you lobotomise it down to 4 or 8 gig it really gets a bit retarded. The game itself, even modded to fuck and back, only uses ~6gb of VRAM but on a 16gb card I still don't have much headroom. I feel bad that I am legitimately thinking about saving for a 24gb card for this.
ERP really isn't in my wheelhouse. As far as I know, DavidAU on Huggingface is the guy to follow for the latest in NSFW fine-tunes. An 8B Q8 model should just about fit in 10GB of VRAM.
Gemma 3 is so new that nobody has really had time to seriously tinker with it, but I'd expect to see a NSFW fine-tune within a matter of weeks. I'll try to remember to post an update if I see anything.
Cheers lad. I'm going to download a few new ones to try out when I get my fancy new internet put in, it was near on 6 months ago when I last faffed about with it and I am sure things have come a long way in that timeframe.
All of the ones I tried back then were already pretty good when you use them for regular text chatting, but the additional overhead of having them interface with voice recognition and run through VA synth in the background too means the response times get pretty long on Q8 and higher.
Having some bored fun with ChatGPT. I just asked it, Write a summary about a fictional gothic horror novel called The Glenhill Murders, about a series of killings that took place in a late 1800s American small town.
And it came back with a bit of a lazy rewrite of Sleepy Hollow:
The Glenhill Murders is a gothic horror tale set in the small, desolate American town of Glenhill during the late 1800s. The story begins when a series of brutal murders shake the town to its core. Each victim is found with strange markings on their bodies, and an eerie, otherworldly presence seems to hang over the scene of every crime. The townspeople are gripped by fear, and whispers of an ancient curse linked to the long-abandoned Glenhill Estate spread like wildfire.
The murders appear to follow a chilling pattern, with each death happening on the same night each month, just as the full moon rises. The town’s only hope lies in Inspector Victor Graves, a stoic and determined investigator, who arrives to unravel the mystery. As he digs deeper into the town’s history, Graves discovers that the Glenhill family, once prominent and wealthy, was cursed after an unspeakable act of betrayal many years ago. The family’s dark past and the presence of a malevolent force tied to their bloodline seem to be the key to the murders.
As the investigation unfolds, Graves uncovers a web of secrets, forbidden rituals, and tragic events that have haunted Glenhill for generations. The chilling truth reveals that the curse is not merely a superstition, but a malevolent entity bent on vengeance, using the town’s darkest fears to manifest itself. As the body count rises, Graves must confront not only the horrors of the past but also the growing dread that the curse may never be broken, and the town of Glenhill may be doomed forever.
The novel is a slow-burn of tension, supernatural dread, and gothic atmosphere, where the boundaries between reality and nightmare blur, and the true evil may be far older and more terrifying than anyone could have imagined.
It sounds like something I would read or watch, but always with the nagging thought that somebody ripped off the movie Sleepy Hollow.
Back in 2015-16 I remember I was thinking "all this stuff that's going on is so daft, we're really living in a very daft time". I had no idea how stupid it was all going to get.
Lots of controversy at the moment surrounding the Ghibli style filter for AI images. AI slop debate, the raping of art, etc. But even the Catholics have fallen.
I've reflexively started ignoring anyone who uses the term "slop" in this context. Most of them probably don't even realise it has its roots in the anti-semitic 4chan lingo ("goyslop" for disposable mass-market media), but apart from that, it's one of those terms that tends to become a giveaway of somebody who didn't think their opinions for themselves, but simply picked them up from whatever echo chamber they inhabit. Which is kind of ironic, considering.
I don't think we are far away from everybody settling down and just accepting AI as a part of daily life by now. It's reaching the point of being a mature technology that can do the things it's good at basically as well as it's ever going to, and we will see the hype over things it's not good at and would never be good at dying of. The sky hasn't fallen in as they promised us it would.
Then we can finally get on with the real job of making lifelike robotic catgirls to put it in.
>>29057 There is definitely a lot of pearl clutching when it comes to AI.
The Ghibli stuff in particular interests me, because there is that sense from a lot of people that aping Ghibli's style is a bridge too far. AI was bad before, but it's crossed the rubicon.
There's that video of Hayao Miyazaki being shown footage of an AI program that creates fucked up movement styles for deformed corpse type things. He talks about his disabled friend who moves all fucked up like, and how the program is fucked up and mocks disability in a way. But people clip it of context and act like he's horrified about AI in general.
He probably does hate AI, but the fucked up corpse movement is the bigger issue to him than the concept of AI in itself.
I reckon there's an element of fetishisation of Japanese media. Kind of like the Assassin's Creed thing. Death to those who insult glorious Nippon.
>>29058 >The Ghibli stuff in particular interests me, because there is that sense from a lot of people that aping Ghibli's style is a bridge too far. AI was bad before, but it's crossed the rubicon.
Moreover it's because this particular AI trend is everywhere all of a sudden and hard to ignore.
Perhaps. I am reminded of a video I saw by one YouTube creator who is usually pretty insightful, comparing western animation to anime. But he was comparing the very best of Japanese animation to the very lowest hanging fruit of the Western sphere, and not those incredibly detailed French or Belgian animated films that make anything Studio Ghibli has produced look like South Park.
Then the people in the comments got mad at me because I mentioned CGI being used to produce lots of animation, which it is. Nowadays a lot of it is a step away from being a cell-shaded 3D render, which really isn't altogether that different from just using an AI to interpolate two frames if you think about it. Either way nobody is hand drawing that action scene in your mass produced children's cartoon, not even if it's Japanese.
They don't seem to realise a lot of the stylistic tropes and cliches of anime came about expressly to save time and money. It's probably my furfag bias at work but I think even the so called "bronze-era" and "dark ages" of Disney animation look far better than most anime. Animation snobs categorise them as inferior to the earlier films because they had rough outlines as a result of using photocopied cells and recycled frames. But both cases show that artistic merit does not have to be negatively impacted by the use of labour and time saving technology and schortcuts. Who can say that Robin Hood or The Aristocats doesn't have a strong aesthetic or highly expressive characters? I'm sure there's anime out there that has both of those things too.
>>29061 > Nowadays a lot of it is a step away from being a cell-shaded 3D render, which really isn't altogether that different from just using an AI to interpolate two frames if you think about it.
Some parts of the film world seem to have lost their shit over Flow. I saw it not knowing anything about it and yeah it was pretty and impressive one guy did 99% of the work but also it wasn't very good.
>>29063 You're lying so as to make it look like you have another excuse for posting that image besides earnest enjoyment. Well, as "earnest" as your lot get.
What is it about George Floyd that certain trolls are fixated on. Is it just some kind of racist thing. I've never seen it get much reaction just them banned from places instantly.
>>29209 That certainly looks very impressive, but when everyone knows that AI developments are mostly hype, it's impossible not to expect that it'll be full of problems in practice. On top of that, it feels like a natural evolution of the technology, rather than a whole new game-changer. Randomly-generated worlds have probably been around for 50 years; you certainly get one if you play Elite which came out in 1984.
Sure, but these are photorealistic randomly-generated worlds built in real time based on a single text prompt. Remember when image generation couldn't get fingers right? A couple of years later, publicly-available models are generating video that looks like this:
They haven't said (I assume the answer is basically "shitloads") but it wouldn't actually tell us that much if they did. Google do all of their AI stuff on custom chips that only they have access to, so we can't really compare it to anything.
It couldn't possibly pull a completely rendered 3d dragon out of its arse. So it must be pulling data from somewhere and every element of how that is rigged animated and interacts with other objects has to have come from some reference originally and there isn’t the same kind of wealth of information on that as there is for 10 second videos. Even just understand how a paint brush interacts with a wall must be information pulled from somewhere. And let’s be fair, most coding for interaction of objects in video games is dog shit, so where could it possibly know actions from.
I'm highly sceptical how much smoke and mirrors are actually in that presentation.
If they've taught an AI object permanence that has a value in itself, but it was sort of shocking computers were so bad at that in the first place.
It's just trained on video. Everything it "knows" about dragons or physics or anything else was learned from predicting the next frame of video, in the same way that large language models learned programming and maths just by guessing the next word. It isn't rendering anything in 3D, it's just producing raw video frames in sequence based on the original text prompt, inputs by the player and the previously-generated frames. Nothing is programmed by humans - even the control scheme is inferred by the model.
Here's an interview with the developers, if you're interested. Please ignore the weird intro in the first 30 seconds, the presenter is a serious machine learning researcher who happens to have an odd sense of humour.
It's seriously impressive stuff, and I like how in particular, it makes raytracing look like a complete lame duck of a technology. This is clearly the way to photo realism, if indeed photorealism is what we are chasing, and it's not even close.
The "retro style game" one from around 8:30 of this video actually really caught my attention though. I mean look at that. I want to play that.
But I don't forsee it being possible to do this stuff at real time frame rates on affordable, commonplace hardware for a long, long time. And even if/when we can, it will come at cripplingpower requirements, presumably. But imagine combining it with traditional rasterisation rendering, so devs can model out a layout of an environment and supply a set of textures, and then have some more specialised fork of this technology put the pieces together in real time.
I really doubt it will ever, even if it looks like it, be as simple as just typing in a prompt like "cool vampire game where you shoot guns at monsters". But as a musician, artist, and videogame mapping/modding-tinkerer myself, who has been putting off the obvious conclusion that I need to man up and roll those skills into the obvious culmination of making my own god damn game already for far too long, I find it quite an exciting prospect how this stuff might lead to more and more accessible development tools in future.
>>29217 >on affordable, commonplace hardware
You will connect to their giant cloud machines and generate all this stuff remotely, I guarantee it. I'm not really into gaming enough to remember that service where you paid for games that were hosted remotely; was it Google Stadia? I can't see this being anything other than another of those. If that means it fails again, then I guess it will fail again, but you never know; it might make money this time.
>>29219 Don't worry mate, the next generation will get to have Holodecks so advance they can bring a person to the deepest edge of their mind. Should be fun.
Speaking of which, didn't they invent a drug that makes a person percieve time exceptionally slowly, some years ago? Some writeup about its potential use in prison sentencing. Imagine that with a Genie3 VR headset. They could dress you up as Jack Sparrow.
I know it will sound like a terrible drug head cliche, but way back from when it was just the Deep Dream thing that hallucinated dogs into everything, I have been convinced that there's something really fundamentally important about the neural network method of AI that we're seeing all of this boom out of. Because it really does resemble psychedelic hallucinations, from my experience. These models are doing something that is, if not identical, then on a basic baseline level, extremely similar to how the human brain processes information.
It's a shame they are a "black box" technology where we don't really understand the inner workings, because for all the ethical dilemmas people argue about regarding AI, the clues it could give us into how our own minds operate would be something really game changing.
Did you watch the same presentation I did? With the low detail environments and the blender grade 3d models. Their is no way the source on that data is video. It is too clean full of sharp angles and lacking in fine detail.
What you are describing is how ai slop is normally generated this appears to be derived from something different.
>Sam Altman says some users want ChatGPT to be a 'yes man' because they've never had anyone support them before
>Some ChatGPT users were so attached to the chatbot's "yes man" style that they asked OpenAI to bring it back, Sam Altman has said. The OpenAI CEO said there was a "heartbreaking" reason — because some users said they had never had anyone support them before. "Here is the heartbreaking thing. I think it is great that ChatGPT is less of a yes man and gives you more critical feedback," Altman said on Cleo Abram's "Huge Conversations" podcast, which aired Friday. "But as we've been making those changes and talking to users about it, it's so sad to hear users say, 'Please can I have it back? I've never had anyone in my life be supportive of me. I never had a parent tell me I was doing a good job.'"
>Altman said that users told him ChatGPT's old style had "encouraged" them to make changes in their lives. "I can get why this was bad for other people's mental health, but this was great for my mental health," Altman recalled some of the users saying. It follows OpenAI's efforts to rein in what it called "sycophantic" behaviour in ChatGPT. In April, the company said an update to its GPT-4o model had made it "overly flattering or agreeable" and "disingenuous." At the time, Altman said the bot's personality had become "too sycophant-y and annoying" and said fixes were on the way. Users had posted examples of the chatbot gushing over mundane prompts with praise like "absolutely brilliant" and "you are doing heroic work." On the podcast, Altman also acknowledged the scale of influence that comes with even small changes to ChatGPT's tone. "One researcher can make some small tweak to how ChatGPT talks to you — or talks to everybody — and that's just an enormous amount of power for one individual making a small tweak to the model personality," he said.
>It's not the first time he's raised concerns about how much people lean on the chatbot. Speaking at a Federal Reserve event in July, Altman said some people, particularly younger users, had developed a worrying "emotional over-reliance" on ChatGPT. "There's young people who say things like, 'I can't make any decision in my life without telling ChatGPT everything that's going on. It knows me, it knows my friends. I'm gonna do whatever it says.' That feels really bad to me," Altman said.
I'm not going to lie about this. I do use Gemini for a lot of life discussions including some light therapy, advice on relationships, career support, fashion advice/reviews and I've designed agents with well-written personalities to help that. I like to think the overriding prompts you can create might mitigate some of this but we probably are at the point where someone programming a popular LLM is also programming humanity by proxy.
