What a silly article. 700,000 per day is ~256 million a year. Thats peanuts compared to the 10 billion they got from MS. With no new funding they could run for about a decade & this is one of the most promising new technologies in years. MS would never let the company fail due to lack of funding, its basically MS’s LLM play at this point.
Almost every company uses either Google or Microsoft Office products and we already know that they’re working on an AI offering/solution for O365 integration, they can see the writing on the wall here and are going to profit massively as they include it in their E5 license structure or invent a new one that includes AI. Then they’ll recoup that investment in months.
While title is click bite, they do say right at the beginning:
*Right now, it is pulling through only because of Microsoft’s $10 billion funding *
Pretty hard to miss, and than they go to explain their point, which might be wrong, but still stands. 700k i only one model, there are others and making new ones and running the company. It is easy over 1B a year without making profit. Still not significant since people will pour money into it even after those 10B.
Openai biggest spending is infrastructure, Whis is rented from… Microsoft. Even if the company fold, they will have given back to Microsoft most of the money invested
MS is basically getting a ton of equity in exchange for cloud credits. That’s a ridiculously good deal for MS.
When you get articles like this, the first thing you should ask is “Who the fuck is Firstpost?”
Yeah where the hell do these posters find these articles anyway? It’s always from blogs that repost stuff from somewhere else
The difference is in who gets the ad money.
I mean, you’re correct in the sense Microsoft basically owns their ass at this point, and that Microsoft doesn’t care if they make a loss because it’s sitting on a mountain of cash. So one way or another Microsoft is getting something cool out of it. But at the same time it’s still true that OpenAI’s business plan was unsustainable hyped hogwash.
Also, their biggest expenses are cloud expenses, and they use the MS cloud, so that basically means that Microsoft is getting a ton of equity in a hot startup in exchange for cloud credits which is a ridiculously good deal for MS. Zero chance MS would let them fail.
Their business plan got Microsoft to drop 10 billion dollars on them.
None of my shitty plans have pulled that off.
If they got any of that into their own pockets kudos to them.
Mainly they used it to pay for the tech and research and it’s all reverting back to Microsoft eventually. Going bankrupt is not quite the same as being acquired.
The thing about all GPT models is that they’re based on the frequency of the word to determine its usage. Which means the only way to get good results is if it’s running on cutting edge equipment designed specifically for that job, while being almost a TB in size. Meanwhile, Diffusion models are only GB and run on the GPU but still produce masterpieces because they already know what that word is associated with.
Would help if they would offer more payment options than just credit card, which is not really popular in many countries
That would explain why ChatGPT started regurgitating cookie-cutter garbage responses more often than usual a few months after launch. It really started feeling more like a chatbot lately, it almost felt talking to a human 6 months ago.
I don’t think it does. I doubt it is purely a cost issue. Microsoft is going to throw billions at OpenAI, no problem.
What has happened, based on the info we get from the company, is that they keep tweaking their algorithms in response to how people use them. ChatGPT was amazing at first. But it would also easily tell you how to murder someone and get away with it, create a plausible sounding weapon of mass destruction, coerce you into weird relationships, and basically anything else it wasn’t supposed to do.
I’ve noticed it has become worse at rubber ducking non-trivial coding prompts. I’ve noticed that my juniors have a hell of a time functioning without access to it, and they’d rather ask questions of seniors rather than try to find information our solutions themselves, replacing chatbots with Sr devs essentially.
A good tool for getting people on ramped if they’ve never coded before, and maybe for rubber ducking in my experience. But far too volatile for consistent work. Especially with a Blackbox of a company constantly hampering its outputs.
As a Sr. Dev, I’m always floored by stories of people trying to integrate chatGPT into their development workflow.
It’s not a truth machine. It has no conception of correctness. It’s designed to make responses that look correct.
Would you hire a dev with no comprehension of the task, who can not reliably communicate what their code does, can not be tasked with finding and fixing their own bugs, is incapable of having accountibility, can not be reliably coached, is often wrong and refuses to accept or admit it, can not comprehend PR feedback, and who requires significantly greater scrutiny of their work because it is by explicit design created to look correct?
ChatGPT is by pretty much every metric the exact opposite of what I want from a dev in an enterprise development setting.
Search engines aren’t truth machines either. StackOverflow reputation is not a truth machine either. These are all tools to use. Blind trust in any of them is incorrect. I get your point, I really do, but it’s just as foolish as believing everyone using StackOverflow just copies and pastes the top rated answer into their code and commits it without testing then calls it a day. Part of mentoring junior devs is enabling them to be good problem solvers, not just solving their problems. Showing them how to properly use these tools and how to validate things is what you should be doing, not just giving them a solution.
I agree with everything you just said, but i think that without greater context it’s maybe still unclear to some why I still place chatGPT in a league of it’s own.
I guess I’m maybe some kind of relic from a bygone era, because tbh I just can’t relate to the “I copied and pasted this from stack overflow and it just worked” memes. Maybe I underestimate how many people in the industry are that fundamentally different from how we work.
Google is not for obtaining code snippets. It’s for finding docs, for troubleshooting error messages, etc.
If you have like… Design or patterning questions, bring that to the team. We’ll run through it together with the benefits of having the contextual knowledge of our problem domain, internal code references, and our deployment architecture. We’ll all come out of the conversation smarter, and we’re less likely to end up needing to make avoidable pivots later on.
The additional time required to validate a chatGPT generated piece of code could have instead been spent invested in the dev to just do it right and to properly fit within our context the first time, and the dev will be smarter for it and that investment in the dev will pay out every moment forward.
I guess I see your point. I haven’t asked ChatGPT to generate code and tried to use it except for once ages ago but even then I didn’t really check it and it was a niche piece of software without many examples online.
Don’t underestimate C levels who read a Bloomberg article about AI to try and run their entire company off of it…then wonder why everything is on fire.
Honestly once ChatGPT started giving answers that consistently don’t work I just started googling stuff again because it was quicker and easier than getting the AI to regurgitate stack overflow answers.
removed by mod
Would you hire a dev with no comprehension of the task, who can not reliably communicate what their code does, can not be tasked with finding and fixing their own bugs, is incapable of having accountibility, can not be reliably coached, is often wrong and refuses to accept or admit it, can not comprehend PR feedback, and who requires significantly greater scrutiny of their work because it is by explicit design created to look correct?
Not me, but my boss would… wait a minute…
Copilot is pretty amazing for day to day coding, although I wonder if a junior dev might get led astray with some of its bad ideas, or too dependent on it in general.
Edit: shit, maybe I’m too dependent on it.
I’m also having a good time with copilot
Considering asking my company to pay for the subscription as I can justify that it’s worth it.
Yes many times it is wrong but even if it it’s only 80% correct at least I get a suggestion on how to solve an issue. Many times it suggest a function and the code snippet has something missing but I can easily fix it or improve it. Without I would probably not know about that function at all.
I also want to start using it for documentation and unit tests. I think there it’s where it will really be useful.
Btw if you aren’t in the chat beta I really recommend it
Just started using it for documentation, really impressed so far. Produced better docstrings for my functions than I ever do in a fraction of the time. So far all valid, thorough and on point. I’m looking forward to asking it to help write unit tests.
it honestly seems better suited for those tasks because it really doesn’t need to know anything that you’d have to tell it otherwise.
The code is already there, so it can get literally all the info that it needs, and it is quite good at grasping what the function does, even if sometimes it lacks the context of the why. But that’s not relevant for unit tests, and for documentation that’s where the user comes in. It’s also why it’s called copilot, you still make the decisions.
I am unsure about the free version, but I really am very surprised by how good the paid version with the code interpreter has gotten in the last 4-6weeks. Feels like I have a c# syntax guru on 24/7 access. Used to make lots of mistakes a couple months ago, but rarely does now and if it does it almost always fixes in in the next code edit. It has saved my untold hours.
Link?
https://openai.com. You have to pay to get the code interpreter as it is part of the plus access. Worth it for me.
Thanks. I’ll have a look
But what did they expect would happen, that more people would subscribe to pro? In the beginning I thought they just wanted to survey-farm usage to figure out what the most popular use cases were and then sell that information or repackage use-cases as an individual added-value service.
Does it feel like these “game changing” techs have lives that are accelerating? Like there’s the dot com bubble of a decade or so, the NFT craze that lasted a few years, and now AI that’s not been a year.
The Internet is concentrating and getting worse because of it, inundated with ads and bots and bots who make ads and ads for bots, and being existentially threatened by Google’s DRM scheme. NFTs have become a joke, and the vast majority of crypto is not far behind. How long can we play with this new toy? Its lead paint is already peeling.
I read an article about the bot collapse. Basically companies use bot to buy add space on websties. Google uses a bot to match adds to websites. Now we have a massive influx of AI made pages. Literally pages of bs just to make more add space that a bot will sell to another not. It is bots all the way down.
Here is an alternative Piped link(s): https://piped.video/uxviBxaSjyc
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I’m open-source, check me out at GitHub.
Good bot.
This AI craze is actually a crypto craze in disguise https://apnews.com/article/worldcoin-cryptocurrency-sam-altman-data-privacy-9dc6a68590435b2f10fedaa0db58331b
As for the pace, I think the US financial services industry has been on a growth spree for decades and they’re desperate to find the new thing that will make them money. It’s like ed edd & eddy but with the PC, internet, dotcom, internet service, social media and now crypto
What?
That has nothing to do with chatgpt or other ai…
shilling
Sam altman, CEO of OpenAI, has nothing to do with chatgpt?
Cryptocurrency has nothing to do with AI. They are completely different technologies.
I never said they weren’t lmao, LLM tech =/= the hype around them, including you
Wtf are you trying to say?
The AI craze is much further reaching than an crypto. Reducing it to just crypto is way too simple.
I wouldn’t put nfts in the same boat as the dotcom bust. The dotcom thing was way bigger. Most people didn’t do anything with nfts. Crypto seems in between. The AI thing seems similar though.
And don’t forget the metaverse!
I mean apart from the fact it’s not sourced or whatever, it’s standard practice for these tech companies to run a massive loss for years while basically giving their product away for free (which is why you can use openAI with minimal if any costs, even at scale).
Once everyone’s using your product over competitors who couldn’t afford to outlast your own venture capitalists, you can turn the price up and rake in cash since you’re the biggest player in the market.
It’s just Uber’s business model.
The difference is that the VC bubble has mostly ended. There isn’t “free money” to keep throwing at a problem post-pan. That’s why there’s an increased focus on Uber (and others) making a profit.
This is what caused spez at Reddit and Musk at Twitter to go into desperation mode and start flipping tables over. Their investors are starting to want results now, not sometime in the distant future.
I don’t know anything about anything, but part of me suspects that lots of good funding is still out there, it’s just being used more quietly and more scrupulously, & not being thrown at the first microdosing tech wanker with a great elevator pitch on how they’re going to make “the Tesla of dental floss”.
In this case, Microsoft owns 49% of OpenAI, so they’re the ones subsidizing it. They can also offer at-cost hosting and in-roads into enterprise sales. Probably a better deal at this point than VC cash.
Speaking of Uber, I believe it turned a profit the first time this year. That is, it never made any profit since its creation in whenever it was created.
All it’s every done is rob from it’s employees so it can give money to stockholders. Just like every corporation.
Whatever, it was always some fly by night operation anyway. It’s a cool toy but all this revolutionary crap talk was just like when NFT’s showed up.
well I mean, chatGPT actually does have some real world use. personally, I find chatGPT more helpful than Stack Overflow when it comes to finding problems with my code
huh, so with the 10bn from Microsoft they should be good for… just over 30 years!
ChatGPT has the potential to make Bing relevant and unseat Google. No way Microsoft pulls funding. Sure, they might screw it up, but they’ll absolutely keep throwing cash at it.
They seems to be killing Cortana… So I expect a new assistant at least based partially on this tbh.
It’s definitely become a part of a lot of people’s workflows. I don’t think OpenAI can die. But the need of the hour is to find a way to improve efficiency multifold. This will make it cheaper, more powerful and more accessible
At a $250mm/yr burn rate and a revenue of… a lot less than that, they can die pretty quickly
Agreed. But it will be a significant loss for a big chunk of people since other LLMs aren’t nearly as good as GPT-4.
That’s fine, I don’t care that there is a good LLM
I think they’re just trying to get people hooked, and then they’ll start charging for it. It even says at the bottom of the page when you’re in a chat:
Free Research Preview. ChatGPT may produce inaccurate information about people, places, or facts. ChatGPT August 3 Version
I don’t think it’s at all clear that that’s a viable business strategy in a market where that kind of sleight of hand is as well known as it is right now
Except for the fact that they’ve said for the entire existence of chatgpt that it’s a free research preview.
Well, I was happily paying them to lewd up the chatbots, but then they emailed me telling me to stop. I guess they don’t want my money.
I’ve heard it’s not as good, but I think NovelAI works pretty well, and explicitly allows that.
I actually started my journey into Lewd AI stuff with NovelAI. I stopped using it after awhile because I like chatbot rp specifically, not just something that will finish a story for me. Using Silly Tavern to try and emulate the NovelAI models into acting like chat bots just shows how not good they are at that.
Makes sense. In that case I guess your next best option is probably to buy or rent hardware to run the local models that are suitable for chat rp.
I have definitely been considering it. My current hardware gives me about an 80 second delay when I run an llm locally.
Same, at least for anything but the tiny ones that will fit in my limited vram. Hoping a gpu that’s actually good for LLM will come out in the next few years that’s not 15k and made for servers.
This article is dumb as shit
No sources and even given their numbers they could continue running chatgpt for another 30 years. I doubt they’re anywhere near a net profit but they’re far from bankruptcy.
The flow of the writing style felt kinda off, like someone was speaking really fast spewing random trivia and leaving
Right!? I believe it has the hallmark repetitive blandness indicating AI wrote it (because oroboros)
If ChatGPT only costs $700k to run per day and they have a $10b war-chest, assuming there were no other overhead/development costs, OpenAI could run ChatGPT for 39 years. I’m not saying the premise of the article is flawed, but seeing as those are the only 2 relevant data points that they presented in this (honestly poorly written) article, I’m more than a little dubious.
But, as a thought experiment, let’s say there’s some truth to the claim that they’re burning through their stack of money in just one year. If things get too dire, Microsoft will just buy 51% or more of OpenAI (they’re going to be at 49% anyway after the $10b deal), take controlling interest, and figure out a way to make it profitable.
What’s most likely going to happen is OpenAI is going to continue finding ways to cut costs like caching common query responses for free users (and possibly even entire conversations, assuming they get some common follow-up responses). They’ll likely iterate on their infrastructure and cut costs for running new queries. Then they’ll charge enough for their APIs to start making a lot of money. Needless to say, I do not see OpenAI going bankrupt next year. I think they’re going to be profitable within 5-10 years. Microsoft is not dumb and they will not let OpenAI fail.
Details of the 10b aren’t public, but we know it’s a multi year deal, so it’s possible that OpenAI doesn’t actually have the full amount in cash now, and they could go bankrupt before they unlock the full amount. In the event of a bankruptcy, Microsoft could be in a position to acquire their assets for themselves on the cheap.
That’s a lot of crypto coins to sell
This article has been flagged on HN for being clickbait garbage.
Flagged where?
I was also curious about that, chatgpt says hacker news and that makes sense to me.
Hacker News
HN is biased towards AI though so 🤷
It is clearly no sense. But it satisfies the irrational needs of the masses to hate on AI.
Tbf I have no idea why. Why do people hate a extremely clever family of mathematical methods, which highlights the brilliance of human minds. But here we are. Casually shitting on one of the highest peak humanity has ever reached
People are scared because it will make consolidation of power much easier, and make many of the comfyer jobs irrelevant. You can’t strike for better wages when your employer is already trying to get rid of you.
The idealist solution is UBI but that will never work in a country where corporations have a stranglehold on the means of production.
Hunger shouldn’t be a problem in a world where we produce more food with less labor than anytime in history, but it still is, because everything must have a monetary value, and not everyone can pay enough to be worth feeding.
I agree with this. People should fight to democratize AI, public model, public data, public fair research. And should fight misuse of it from business schools’ type of guys.
I probably sound like I hate it, but I’m just giving my annual “this new tech isn’t the miracle it’s being sold as” warning, before I go back to charging folks good money to clean up the mess they made going “all in” on the last one.
Because it’s just the same as autocomplete on your phone lol so whatevs.
/s
It seems to be a common thing. I gave up on /r/futurology and /r/technology over on Reddit long ago because it was filled with an endless stream of links to cool new things with comment sections filled with nothing but negativity about those cool new things. Even /r/singularity is drifting that way. And so it is here on the Fediverse too, the various “technology” communities are attracting a similar userbase.
Sure, not everything pans out. But that’s no excuse for making all of these communities into reflections of /r/nothingeverhappens. Technology does change, sometimes in revolutionary ways. It’d be nice if there was a community that was more upbeat about that.
If AI was so great, it would find a solution to operate at fraction of the cost it does now
Wait, has anybody bothered to ask AI how to fix itself? How much Avocado testing does it do? Can AI pull itself up by its own boot partition, or does it expect the administrator to just give it everything?
It’s working on it. It’s just a pretty difficult task to be better than 50 years of optimization by human computer scientists https://www.nature.com/articles/d41586-023-01883-4
I asked Bard how ChatGPT can fix itself, here’s what it said:
ChatGPT can fix itself in a number of ways, including:
Retraining on a larger dataset. ChatGPT is trained on a massive dataset of text and code, but it can always be retrained on a larger dataset. This would help ChatGPT to learn new things and improve its performance.
Using a different algorithm. ChatGPT is currently trained using the GPT-3 algorithm, but there are other algorithms that could be used. For example, ChatGPT could be trained using the Megatron-Turing NLG algorithm, which is one of the most powerful language models in the world.
Using a combination of algorithms. ChatGPT could also be trained using a combination of algorithms. This would allow ChatGPT to benefit from the strengths of different algorithms.
Using a feedback loop. ChatGPT could be used to generate text, and then this text could be used to train ChatGPT. This would allow ChatGPT to learn from its own mistakes and improve its performance over time.
Using human feedback. ChatGPT could also be used to generate text, and then this text could be reviewed by humans. The human feedback could then be used to improve ChatGPT’s performance.
“Using a feedback loop. ChatGPT could be used to generate text, and then this text could be used to train ChatGPT. This would allow ChatGPT to learn from its own mistakes and improve its performance over time.”
So basically create its own Fox News and see how that goes.
this is widely known to very quickly destroy your model
Remember, this is Bard’s advice for ChatGPT
The feedback loop is already happening, and is called model collapse.
It’s not a good thing.
The full suggestion includes “This would allow ChatGPT to learn from its own mistakes”, which implies that the text it generated would be evaluated and curated before being sent back into it for training. That, as well as including non-AI-generated text along with the AI generated stuff, should stop model collapse.
Model collapse is basically inbreeding, with similar causes and similar solutions. A little inbreeding is not inherently bad, indeed it’s used frequently when you’re trying to breed an organism to have specific desirable characteristics.
If having an AI tell researchers that they should base its next iteration off of Megatron isn’t the plot of a Michael Bay Transformers movie already, it should have been.
Really says something that none of your responses yet seem to have caught that this was a joke.
OP might have been intending it as a joke, but self-improvement is a very real subject of AI research so if that’s the case he accidentally said something about a serious topic.
It’s an essential part of the idea of the technological singularity. An AI iterates itself and the systems it runs on, becoming more efficient, powerful, and effective at a rate that makes all of human progress up to that point look like nothing.
While I’m inclined to believe the singularity is achievable, it’s important to remember that there’s no evidence today that it will ever be reached.
Our hope for it, and the good than can come with it, can’t pull it into the realm of things we will see in our lifetimes. It could emerge soon, but it’s at least as likely to stay science fiction for another millennia.
Yeah, when chat gpt 4 first came out, I thought we might be close. But as it’s capabilities and limitations became more clear, it doesn’t look like we’re close at all. I mean, it’s hard to say for sure since an LLM will just make up a part of an AI and maybe the other pieces are farther along but just not getting as much attention because there’s value in not making those things public.
But as someone who works in one of the fields that would be involved in the technological singularity, no one really knows good ways to apply AI to the work we do and the best initiatives I’ve seen come out of the corporate drive to leverage AI aren’t actually AI, but just smarter automation tools.
The upvotes vs comments shows the vocal minority is just doing vocal minority things.
It doesn’t “know” anything. It can’t solve that problem. It’s trained on humans so it’s limited to what we have written down.
I love ChatGPT but if it’s creative it’s because you asked it the right questions and found an oblique answer yourself.
if we don’t know, it doesn’t know.
If we know, but there’s no public text about it, it doesn’t know either.
it is trained off of stuff that has already been written, and trained to emulate the statistical properties of those words. It cannot and will not tell us anything new
That’s not true. These models aren’t just regurgitating text that they were trained on. They learn the patterns and concepts in that text, and they’re able to use those to infer things that weren’t explicitly present in the training data.
I read recently about some researchers who were experimenting with ChatGPT’s ability to do basic arithmetic. It’s not great at it, but it’s definitely figured out some techniques that allow it to answer math problems that were not in its training set. It gets them wrong sometimes, but it’s like a human doing math in its head rather than a calculator using rigorous algorithms so that’s to be expected.
they learn statistical correlations between words. given the last 5000 (or however large the context is) words, and absolutely no other information besides that, what is the most likely word to appear next? It’s a glorified order 5000 markov chain.
The reason it can “do” some math is that there are tons of examples in the training set using small numbers usually used as examples. it can do basic arithmetic because it has seen “2+2=4” and other examples with simple numbers like that. The studies used test basic arithmetic. The same things that it had millions of pre-worked examples of. And it still gets those wrong, with astonishing frequency. those studies aren’t talking about asking it “what is the square root of pi” or stuff like that. but stuff such as “is 7 greater than 4?”, “what is 10 + 3?”, “is 97 prime?” stuff it has most definitely seen the answers to. ask it about some large prime, and it’ll nay no, and be probably right, because most numbers are composite
those studies aren’t talking about asking it “what is the square root of pi” or stuff like that. but stuff such as “is 7 greater than 4?”, “what is 10 + 3?”, “is 97 prime?” stuff it has most definitely seen the answers to.
No, they very explicitly checked to see whether the training set contains the literal math problem that they asked it for the answer to. ChatGPT is able to answer math questions that it has never seen before. I believe this is the article (though I had to go searching, it’s been a while).
When people dismiss LLMs as “just prediction engines” they’re really missing the point. Of course they’re prediction engines, that’s not in dispute. The question is about how they go about making those predictions. When I show you the string “18 + 10 =” you can predict what comes next, yes? Well, how did you predict it? Did you memorize that particular specific string, or have you developed heuristics for how to do simple addition problems when you see them?
These things are currently infamously bad at math, though.
I won’t argue that it’ll never get there. I’m confident it will, - though with a lot more perl hacks than elegant emergence.
But today, these things have an astonishingly high ‘appearance of intelligence’ to ‘incredible stupidity’ ratio.
Humans are also not particularly well known for their math skills. Ask a random stranger to do simple arithmetic in their head, with only a few seconds to think and no outside help, and I wouldn’t expect particularly reliable results.
It gives me the giggles that folks speculating are getting more upvotes than your post that demonstrates actual understanding of the implementation details.
If I were the type to sell sizzle hype and snake oil, now would be the time to do it. The venture capitalists may have learned their lesson, but the general public haven’t.
Deepmind is actually working on an AI that improve performances of low level programs. It started with improving sorting algorithm.
It’s an RL algorithm.
Main issue is that everything takes time, and expectations on current AI are artificially inflated.
It will reach the point most are discussing now, it’ll simply take a bit longer than people expect