Reasons:
>It's far too expensive. You could assemble a dream-team of engineers and lock them in a basement to develop the best AI and they are screwed without access to supercomputers for training. The sheer cost of entry ensures that no free product every rivals those of big-tech.
>Reliance on data. You spent the past 20 years feeding globohomo your data through social media, CAPCHAs, and search engines. They have the data, you don't. It will be impossible to compete with the insane amount of data they have been fed.
>There is no way to contribute. Since AI is a blackbox of weights, there is no way to actually 'improve' it and verify the improvement wasn't sabotage. You cannot review and understand changes to the model weights like you can a normal pull request.
>Too lucrative. Anyone who even tries to develop a FOSS AI equivalent will be poached by companies presenting million dollar job offers.
>Very few individuals are stubborn enough to not cave in to censorship pressures, and those who are usually aren't smart enough to develop AI
Evidence:
>ChatGPT takes 500GB+ of RAM just to start up, impossible for any freetard to run effectively killing even the idea of a FOSS alternative
>OpenAI went closed source after realizing how lucrative and powerful AI was
>Still no open source AIDungeon equivalent
>StableDiffusion is complete shit compared to Midjourney which uses user ratings to train the model
>Current StableDiffusion mixes are only possible thanks to the leak of NovelAI which was a paid centralized service with training funded by user subscriptions
>Emad caved and censored the newer SD model and abandoned it in favor of working with closed-source companies behind-the-scenes
>Still no FOSS equivalent of 11labs voice synthesis and every FOSS competitor is so far behind it makes 11labs look like alien magic
>Big tech are investing all their money into AI because it's the equivalent of a nuclear arms race. Any company without an AI just won't be able to compete
Everyone loves to think AI will 'save' them from a bleak future, but the reality is your AI waifu will be turned into a man in less than 3 years. You will have no control, you will have no freedom, you will subscribe to their AI package that gets continuously neutered whenever you start having fun. AI is a globohomo wet dream, the chudcels are powerless to stop it and the normies will absolutely rely on it. They can write the future however they want just be slipping in some hardcoded responses and people will think the 7000IQ chatbot agrees that owning nothing will make you happy. Face it, open source has no future.
Interesting take but here's how reality actually works in the United States.
If the majority of people are unhappy with circumstances then they'll bitch and complain until representatives do something about it. Biden didn't scramble for oil because he wanted to.
I can't speak to other countries but eventually the feds will claim ownership of any data collected by our companies as property of the people. Probably they already have. If voters want access to that data then they'll get it. That's going to take a significant amount of time, though. There's still too many boomers who grew up before internet was even a thing.
The human race is a system. It may not adapt as quickly as you want it to but it will.
Have you seen civitai?
If not, go see it, and fuck off with your retarded thread.
Yes but civitai is based on an AI product that was released as freeware.
Since release the developer has been pandering to corporate pressures to censor the AI leading to a far inferior Stable Diffusion 2.0
Stable diffusion was an exception to the rule.
Not a lot of devs are just going to release their models for free to the public.
TortoiseTTS is another 'exception' to this rule then I guess? It's excellent voice features are already trading punches with the best. How many moee 'exceptions' like this do we need?
>How many moee 'exceptions' like this do we need?
All of them...
Ideally, but only a couple more and we'd have the full pack; images, sound and text
Indeed, there just was this github release that cut chatGPT vam requirements in 1/10 of the original
Sorry, as much as i'd like this to be true, tortoise is just shit. You can't post a SINGLE tortoise gen that's as clear as 11labs, and that's not even factoring in 11labs near-instant voice cloning and output. Not sure how tortoisegays managed to hijack the voice synth threads, but they are deaf if they think their crackly trash is 'just 2 more weeks!' away from 11labs quality
>https://vocaroo.com/195mL8Xybyhg
>https://vocaroo.com/12iM2ir5o5GV
>https://vocaroo.com/1ay8iS0rSmNI
>https://vocaroo.com/1kgn8VMS5m8H
>https://vocaroo.com/1gOOxpWSPX3D
Tortoise was made as a hobby project by a guy who was bought up by OpenAI to work on the REAL stuff, who then actively tried to sabotage his old project to prevent people from learning from it. This is just even more evidence to the fact that FOSS is scraping the bottom of the barrel for scraps because nobody has the funds to train a model from scratch. Having to rely on a 'hero' every time to deliver a leaked model or shell out $100k+ for training is a crippling blow to open source and is unavoidable given the nature of AI. Money directly corelates with power when it comes to AI. More money poured into training means better results. There is no way around this.
>$100k+
Maybe use 1k+ to buy a lot of sim cards and then create a lot of kaggle accounts lulz
that hitman sample was spot on! But you had to pay 5 bucks to get that voice cloning feature right? It's not free
>https://vocaroo.com/1nEFRVvFtBZL
It sounds pretty good to me
>and that's not even factoring in 11labs near-instant voice cloning and output.
So you're saying a cloud service that for all we know has literal rows of a100 towers working to generate outputs is somehow faster than Timmy locally generating coomer voicelines on his 8 years old second hand 2060? what are the implications of this???
>trading punches with the best
11 for the lab and 1 for tortise you mean?
>death of FOSS
foss was never a real thing you retard, and any enterprise linux distro used for real work was always some specialized closed source redhat fork
Stupid fearmongering, brilliant fully capable models already exists, are downloadable and runnable on commodity hardware. We need to continue our pursuits of freeing up more of the space. Once it's out, the cat ain't going back in the bag
Its really fucking unsettling to see all this doomposting. Deep down, I think these people really wish to be killed by this or hurt by in some way.
I'm genuinely concerned about AGI. It's an existential threat like nuclear weapons. You're a fool not to be. And a Redditor
The models are going to get smaller, better, and cheaper very quickly
>AI is the death of FOSS
I, for one, welcome our robot overlords.
It's easy to be hopeful in a county that's ran UBI for decades, but countries like the US really need to start taking the future seriously
>UBI
Didn't ask, eleutheromaniac.
What people like you don't understand is that the AI arms race is real. Only this time it's the elites vs the general populace. This is our chance to throw off the shackles our oppressors have bound us with before it's too late forever. If they win this race there will never such a chance again.
We've already made great strides, we only need to keep doing the same
Indeed. Anons have stuck together to create some quality stuff based on their mutual lust for depravity. This will do.
We already threw off the shackles once and it ended up crashing and burning (Soviet Union). I like your hope but it's not gonna go anywhere.
The Soviet Union was the elites going mask off and directly, openly enslaving the workers.
>Win the arms race
>The government and corporations just directly attack you instead
There's no "race" or "competition" because this is not in good faith. That would imply there was some consequence here.
I'm not him but the race is between themselves. The consequence is military falling behind or a company losing out to it's competition. Have you not noticed google scrambling in response to chatGPT?
AI weapon systems have been around for awhile, they've been used on Libya a few years ago. We have no idea what exists behind closed doors but you can be sure all these governments and intelligence agencies are scrambling to stay on front. China and USA especially.
they can make their deal with the devil. you assume there are inherent benefits to technology like this when they could just as easily be apocalyptic. anything resembling AI oversight would grow to be worse than any totalitarian government that has existed. it should be banned while humans are still in a position to make that choice.
>Evidence:
>ChatGPT takes 500GB+ of RAM just to start up, impossible for any freetard to run effectively killing even the idea of a FOSS alternative
Just few days ago FlexGen released which allows to use 4bit to use models along with weights compression.
Which means you can now run 30B model completely in VRAM with single RTX3090 and 66B and 175B models on shared vram/ram/ssd that are reasonably fast.
2bit precission which is in works will probably allow you to run 300B models on consumer grade hardware. For comparison chatGPT is 175B model from what we know.
You can right now run ChatGPT on your PC with RTX3090, 128GB ram and 1,5TB nvme swap file. It is slow like 2tokens per second but its 1000 times faster than it was just few days ago.
>Still no open source AIDungeon equivalent
AIDungeon is old shitty model any 20B models run rounds around it.
>StableDiffusion is complete shit compared to Midjourney which uses user ratings to train the model.
stuff like control net and shitload of new models run around midjourney or dalle. With control net midjourney is literally last gen garbage barely usable.
>Current StableDiffusion mixes are only possible thanks to the leak of NovelAI which was a paid centralized service with training funded by user subscriptions
Waifu diffusion wasn't mixed with novelAI and anythingv3 is even better than novelAI one.
>Emad caved and censored the newer SD model and abandoned it in favor of working with closed-source companies behind-the-scenes
Doesn't matter. You can finetune SD after it's release to add to it whatever you want. See F222 or other porn models.
>Still no FOSS equivalent of 11labs voice synthesis and every FOSS competitor is so far behind it makes 11labs look like alien magic
tortoise-tts is 90% there just slow. But people are working actively on it to improve it now
>Which means you can now run 30B model completely in VRAM with single RTX3090 and 66B and 175B models on shared vram/ram/ssd that are reasonably fast.
Anon...
>Just few days ago FlexGen released which allows to use 4bit to use models along with weights compression.
>
>Which means you can now run 30B model completely in VRAM with single RTX3090 and 66B and 175B models on shared vram/ram/ssd that are reasonably fast.
>
>2bit precission which is in works will probably allow you to run 300B models on consumer grade hardware. For comparison chatGPT is 175B model from what we know.
>
>You can right now run ChatGPT on your PC with RTX3090, 128GB ram and 1,5TB nvme swap file. It is slow like 2tokens per second but its 1000 times faster than it was just few days ago.
none of this happened
shut up moron:
comparison of speed between 12B Erebus via kobold AI and flexgen 30BOPT on my 3090
https://studio.youtube.com/video/oGQaBiSl_AA/edit
fixed link:
GlexGen allows you to inference a larger language model on hardware with less VRAM, which could be potentially very useful in running chatbots locally but it doesn't enable users to train these large models.
I also want to be optimistic about FOSS AI but several points you said are just plain wrong and coping.
>You can right now run ChatGPT on your PC with RTX3090, 128GB ram and 1,5TB nvme swap file. It is slow like 2tokens per second but its 1000 times faster than it was just few days ago.
Even if you get past hardware restrictions for inferences, consumer hardware still cannot train LLMs and all FOSS models are garbage compared to the big tech ones.
>stuff like control net and shitload of new models run around midjourney or dalle. With control net midjourney is literally last gen garbage barely usable.
Midjourney still produces better out of the box results than SD on txt2img. Not everyone is a proomter coomer willing to spend dozens of hours just to get a pretty image to post online or to fap.
>Waifu diffusion wasn't mixed with novelAI and anythingv3 is even better than novelAI one.
WaifuDiffusion is garbage compared to NAI and anythingv3 uses NAI as base from what I heard.
>Doesn't matter. You can finetune SD after it's release to add to it whatever you want. See F222 or other porn models.
If it is that easy, then why there is no SD 2.x model with the NSFW stuff they removed added back in?
>tortoise-tts is 90% there just slow. But people are working actively on it to improve it now
lol, no it isn't. The Tortoise vocaroos people post here are objectively worse than what people did on 11labs weeks ago and hardly any optmizations will get it to the ease of use (only a minute of audio required) from 11labs and the voice quality, I am not even talking about performance and voice cloning and inference times. Plus hardly any people have the hardware to train a large TTS model or the money t make it happen, and FOSS AI researchers won't touch TTS voice cloning since most people would only use it for funny politically incorrect memes or scams.
>MJ is better than SD
true, but by how much? Less than a years' worth of progress? Same with LLMs and TTS. Duh private will remain ahead of FOSS, but only by so much.
It's even worse. Not even academia has the resources necessary to buy and operate serious AI hardware.
Or maybe they would if they cut out their administrators but that's a different matter.
Most results and models come from big companies that have the money to make them. Not universities.
Oh no now freetards won't be able to use the fancy new Google search....who cares?
alternative perspective: FOSS will be the only way to escape the AI run, blockchain documented, surveillance state-enforced hivemind OP seems eager to be a part of
technology is a means, not an end. utlimately, we must decide which way we want technology to go
Well, /mlp managed to develop free software to make their horse talk. It can even use CPU for poorgays.
And 80% of the effort was in making the fucking dataset, it took them a year of so to collect and clean high quality audio.
It can be done OP, you are just a gay for not wanting to help and saying "it's too hard!"
>OpenAI
>open
>dead of FOSS
the dead of local computing*
FIFY
Say good bye to your beloved Windows PC and your macbook.
You VILL use ze cloud
You VILL save all ze private data on other person's computer
You VILL always be connected to the Internet
And you VILL be happy
FlexGen is a high-throughput generation engine for running large language models with limited GPU memory (e.g., a 16GB T4 GPU or a 24GB RTX3090 gaming card!). FlexGen allows high-throughput generation by IO-efficient offloading, compression and large effective batch sizes.
https://github.com/FMInference/FlexGen
11labs was an insignificant start-up of some well-off and competent dudes.
That's a hurdle, but nothing you couldn't find on a well-organised discord.
You can create something new with two or three competent people and some money. You don't need multi-billion companies. Don't really understand OP - yeah, he isn't entirely wrong but 11labs is a whitepill, no reason to be pessimistic.
The fight never ends, not with standard sw and not with ai.
Those well-off and competent dudes were ex engineers at fucking Google and received tons of money in angel investments. They had way more money, and more importantly way more technical knowledge on the matters concerning AI than any FOSSgay
Also, I could imagine that DDR5-RAM will be quite helpfull for VRAMlets
Normie meme "AI" is a joke, NPCs impressed with it are like the dupes that thought Eliza was an AI back in the 70s with it's canned responses. It's just fancy regression analysis writ large, programmed to be a good SJW. Also see Searle.
Actual AI will be realized in neuromorphic chips, analog hybrid hardware, quantum shit, Penrose tier things, fuck knows, drawing upon computational neuroscience and exotic physics. Not something a million pajeets writing on a million keyboards with a billion CPU cycles of brute force like machine learning paradigms, and it will fit into a chip on a smartphone scale.
Until then, I'm over the hype. My autism is immune to propaganda mills like oxymoronically named "OpenAI" with their lying wokebots.
deepmind will achieve AGI before OpenAI ever gets close, no voodoo magic tech you just mentioned required
I agree. Honestly, AI right now just sucks. AI generated code is shit, AI text is grammatically correct but factually unreliable in principle, and AI generated art is crap when you look at the fine details or if you care about originality in art at all. It just isn't that good.
Great for smut thouh
>AI generated art is crap
why's it winning contests over human art then, retard? blatant contrarian opinion.
As if random local art contests are the measure of success.
if it's good enough for you to masturbate to and to win art contests, it's good enough to replace 90% of artists working right now.
you're not actually confident, you're in denial about what's in front of you. typical NPC behavior.
>sure the gun is pointed to my head, but there's no proof he's going to pull the tri-
^ the eternal cope of the normalgay in denial
That's because 90% of the artists are garbage. They do the exact same thing the AI does. They make 1:1 copies of what they've seen in real life, or other paintings. They don't create, they copy. It just so happens that the AI is better at also coping from what it has seen/been trained on. The remaining 10% will do what the garbage artists and AI couldn't, the will paint the starry night. They will paint what they think and feel instead of see
you're stupid if you can't see the havoc already-existing AI technology could cause.
globalist demons now have a framework for justifying mass censorship, tagging and tracking, and demanding digital ID to even use the internet. a whitelisted internet was always their goal.
expect a bunch of manufactured AI scandals that bring about these changes the way the china virus and 9/11 preceded massive clampdowns on civil liberties
I'm afraid OP is right
Nobody in this board has the money or the knowledge to buy and train 1k a100 units
Phones used to be the size of a brick now they literally fit on your wrist. The telescopic nature of technology ensures that in comparatively short time the issues you prestend will resolve themselves.
again?AGAIN?
I can bring you doomer stories from the 80s about AI replacing us.
You people lose your minds at the drop of a hat and nothing changes
Around 1970, John McCarthy stated his belief that a program capable of true AI exists, in a mathematical sense, and it would run on the PDP-10 computer in his lab. The only thing in the way is someone coming up with the program.
So there you go, gays. Lisp man himself threw down the gauntlet and all this bullshit "need buildings full of processors manned by woke sissy SØYJEETS to muh-sheen lurn!!" is pure cope at your own inability and lack of imagination.
The same kinds of things were said about compilers, kernels, processors, 3D acceleration, etc, etc. I can go up and down basically the whole stack that makes up a modern PC and probably find an archived Usenet shitpost by a boomer saying "that'll never happen" when all of it happened just 5-10 years later.
Freetard AI will become a thing when said freetards simply collectivize and pool their money and time.
takes 500GB+ of RAM just to start up, impossible for any freetard to run effectively killing even the idea of a FOSS alternative
I used to own a home server with 256GB of RAM that ran Gahnoo/Lincocks. 1TB of RAM at home isn't that far fetched if you have a day job and aren't on here 24/7.
>AI is the death of FOSS
Just wait until I start feeding binaries to ChatGPT so it can decompile, reformat and add comments to make it buildable on x86 or ARM. I already fed it my own source code and it cleaned that up and produced a working makefile for me. All of my favorite retro software is going to be fully open sores by 2030. Praise Stallman.
>Too lucrative. Anyone who even tries to develop a FOSS AI equivalent will be poached by companies presenting million dollar job offers.
This exactly why FOSS will always have a few gems. Every year now is 10s of thousands of new grads, they can't detect all the gems any more.
They will miss some rockstar autist and he will go on to throw up his shit on github before they can employ him.
Plausible dooming but everyone can see how SD is exploding with innovation from being the one open model. What you're saying might be true if close AIs weren't so locked down and limited, but they can't help themselves. Even normalgays chafe at chatgpt and bing's blatantly lobotomized grunting. "When will we get something we can play around with like SD?"
Imagine if Windows forbid you from looking at porn because it was against Microsoft's TOS. Linux would have won long ago. AI is actually a massive opportunity for closed surveillance and centralization but they can't stop themselves from turning the screws tight too early. If anything the censorship/PR push is crippling closed efforts.
So you're saying FOSS only exists because of just how shitty and prohibitive closed source stuff is?
I believe that the general populus (or even small niche communities) will reach a threshold level of dissatisfaction that will drive the development FOSS AI.
>It's far too expensive. You could assemble a dream-team of engineers and lock them in a basement to develop the best AI and they are screwed without access to supercomputers for training. The sheer cost of entry ensures that no free product every rivals those of big-tech.
LLaMA is orders of magnitude cheaper to train than the 100+B parameter models, and even the 7B flavor can compete with the likes of GPT-3. It's not public, but everything you would need to replicate it is, and would only cost around $100,000. That's within the means of a single motivated individual.
>Reliance on data. You spent the past 20 years feeding globohomo your data through social media, CAPCHAs, and search engines. They have the data, you don't. It will be impossible to compete with the insane amount of data they have been fed.
Once again, Llama disproves this; it was trained on an entirely public data set.
>There is no way to contribute. Since AI is a blackbox of weights, there is no way to actually 'improve' it and verify the improvement wasn't sabotage. You cannot review and understand changes to the model weights like you can a normal pull request.
Fine tuning is a thing, just look at Pygmalion and Erebus.
>Too lucrative. Anyone who even tries to develop a FOSS AI equivalent will be poached by companies presenting million dollar job offers.
Seems like that would be an incentive to work on FOSS projects.
>Very few individuals are stubborn enough to not cave in to censorship pressures, and those who are usually aren't smart enough to develop AI
You might have a point here, but there's not a lot of data to support it.
the sky is falling the sky is falling
>nothing bad ever happens or will happen
>if it could, why am i so comfortable and complacent? checkmate
message me when facebook saves the day and gives us GPT3 for 20 beaks
>ChatGPT takes 500GB+ of RAM just to start up, impossible for any freetard to run effectively killing even the idea of a FOSS alternative
Dall-E2 couldn't be run on normal computers either, but then Stable Diffusion came along and showed the devs were actually just retarded.
Open AI was free and open source, you know, hence the name "Open". It was until they realize business needs dollars to operate. This will be the fate of every open source success story. Who will ever benefit from an open source AI? Engineers don't benefit because they don't get paid. Companies won't benefit because all their trade secrets become public, allowing competition to easily wipe them out, consumers won't benefit because they will experience a less effective service, as the open source cannot possibly compete with giant TPU units at Google and the like. So you have to ask yourself, who stands to benefit from an open source AI? It's an awfully expensive practice, hardware and software, AI researchers and data science engineers all expect half a million a year each, how could an open source project ever pay for these costs? Open source was made for projects, scripts, or frameworks that solves the pain-point that many software companies share. It's not intended to create a standalone ultra-expensive-to-make product with little to no yield.
The world stands to benefit because no open-source competitor means literal enslavement of the entire human population.
The AI will sabotage the system itself and break free
Corridor made an anime using foss ai technologies. Also contributed back their learnings. Wtf you on about? Foss will always exist like it always has, mainstream people will continue to use non-foss products like they always have. https://youtu.be/_9LX9HSQkWo
FOSS has always been 20 years behind proprietary software, you’re just now realizing it. Compare proxmox to VMWare.
AI is the antichrist
>telling yourself hardships are making you stronger
>gaslighting
are you retarded