Google Screws Up + 4 more AI NEWS
HTML-код
- Опубликовано: 29 фев 2024
- Get early access here! bit.ly/LTXsebastiankamph @LTX.Studio
Google Genie sites.google.com/view/genie-2...
Supir supir.xpixel.group/
Google Gemini blog.google/products/gemini/g...
Get early access to videos and help me, support me on Patreon / sebastiankamph
Chat with me in our community discord: / discord
Stable Diffusion for Beginners Playlist • Stable Diffusion Begin...
My Weekly AI Art Challenges • Let's AI Paint - Weekl...
My Stable diffusion workflow to Perfect Images • Revealing my Workflow ...
ControlNet tutorial and install guide • NEW ControlNet for Sta...
Famous Scenes Remade by ControlNet AI • Famous Scenes Remade b... 
Хобби
Awaiting for your SUPIR installation for A1111😊 12gb version
7:05 Not "too diverse", just dysfunctional. The default should be what is normal for that group. If you want an european samurai, this shouldn't show up if you just search "samurai". Same if you want a samurai with purple hair, pink eyes and bear skin, you should have to add those promps spesifically.
It seems like they are doing this but it looks like they are injecting prompts like "but with diverse people" and then "white european" as a negative injection prompt. It's because Google, the 1.7 trillion dollar company, got lazy. They took a pre-existing image model, bundled it with this other LLM product they spent tons of resources building, then some dope at the top told them to "fix it" they way they wanted it to work. (I'm obviously just guessing)
It also seems the way it acts like when you ask for a white person they had told the LLM to treat it as the same category as illegal/taboo images, and refuse to give it to you.
I would say "colorblind".
Agree 100%. It should be up to user to specify what the user wants otherwise you just get a generic image. Especially when it comes to history, there should not be a "diverse" word added in
When you eliminate a specific group of people who share a characteristic, the correct word is not "diversity."
@@dariusz.9119that's the problem with these models, if you ask for "a picture of a human" you are gonna get a white male 99% of the time. It looks to me they were actually trying to find a solution for the biases of these models for ended up over correcting. At leat they are trying...
Speaking of walls - did you hear about the wall which Marylin Monroe walked past?
It got bricked up.
I am so sorry lol
This was the first time ever that I really noticed a great improvement in a RUclips videos editing. It is a great improvement. Looks super slick. Feels very high quality.
That's very kind of you, you're the real mvp!
Programming is not a just about knowning a specific programming language - I've forgotten about as much of these as I've learned in my life - it's about thinking in a structured way, having at least a vague idea about algorithms. being aware of complexity, knowing how to get it under control and finally there's the domain specific understanding that's required too. So no, programmers will not be obsolete any foreseeable future. AI can do boring and time consuming stuff though, including writing scripts to do this or that. Also it's probably not a good idea anyway to rely on AIs to write all programs, build and fix all machines on their own and nobody around understanding what's actually going on.
Honestly, there's a lot of similarities with AI prompting and programming. I could maybe see AI prompting become the next level of abstraction for programming.
@@peterlewis2178 If you can't understand what the machine is doing, prompting alone won't help. For example if I have to repeat every some minutes that I want an early return approach instead of a 10 levels deep nested if-then-else, that just slows me down. Or if I ask for a minor adjustement and large parts of the code are rewritten so I have to check if this is still even valid. Or if I ask for something that is impossible due an oversight or simple typo on my side but the machine tries to implement it anyway instead of asking back or telling me about the issue like a human developer would. Or of the code is just to complex so it's very hard to understand because the KISS principle is apparently unknown to the machine. So fire-and-forget is definitly not there yet.
@@testales Oh, it's definitely not there yet, no debate there. And I think there will always be value in some people still knowing how to use current programming languages, just like how some programming languages give you more control than others. But I could still see AI prompting becoming a more accessible and widespread alternative to a lot of programming languages, and regular programming languages becoming more of a specialist field for specific needs.
Google needs a new, visionary CEO
That will never happen. The woke cancer is already at stage 3 in Google.
SUPIR uses llava and SDXL. I'm waiting for a workflow and custom nodes for ComfyUI doing something like SUPIR
Google AI is lying, that's the problem.
You mean a company that started out by DNS poisoning the internet to steal traffic is lying... say it ain't so.
Just goes to show that AI can be programmed to the programmers' biases. It shows on the chat AIs and now on the image AIs. Goes to show that AIs programmed by these big companies can't be trusted.
Google is lying, Google's AI is the apple not falling far from the tree.
It's exactly what happened to HAL 9000.
@@jtjames79 true
It does what it’s programmed to do blame googles woke policy
I was able to get SUPIR running on an A100 notebook using Colab Pro, and after playing with it quite a bit the last few days I have to say it is pretty amazing. It does the upscaling in 2 stages:
(1) - The first stage eliminates the majority of the noise with a strong style/filter pass. This basically leaves you with a clean latent of sorts.
(2) - This clean latent gets handed over to a conditioner, which sits on top of a base SDXL checkpoint to perform the upscaling, complete with some generic positive & negative prompts.
Easily the best results of any upscaler I've seen, but it's also the most flexible. You can swap in any SDXL model, change the prompts, modify the stage 1 noise/output image, etc. so you have a ton of flexibility as well in case your upscaled image needs some help. I'm very very excited about the possibilities with this kind of upscaling!
Where can you get Supir?
Thanks for the update, appreciate
I'd argue that learning to program with some form of language, even though it becomes "obsolete", is a good pedagogic device in developing a good sense of code engineering, design and problem solving.
For now sure, but in 20 years, probably not.
Imagine a future where nobody knows how to code anymore, because the last people that knew have all died out.
Then imagine the machines starting to break down because of a massive glitch in the AI code and that code was automatically passed to every AI machine out there.
Now you are left with nobody that knows how to fix them and all the machines we depend on no longer work.
This is why we should always teach code, even if it seems pointless. You have to think of all situations, not just the situations where things are working.
We are running towards the future of (H.G. Wells) The Time Machine.
This is already the case in AI software. The programmers do not know why AI makes certain decisions.
@@somedude5951 yeah it's the ultimate black box, where it's too complex to actually understand what is going on. Too many linked connections.
That's basically what is happening in Russia, now. The last really "good" education system they had was during Soviet times. Most of their engineers are in their retirement age. They've lost the know-how to build a lot of their old tech.
You literally just said the plot of the movie Idiocracy.
Legendary Dust 2 cs 1.6 thank you for the content friend
Still my favorite game to date. The many days I spent playing that game...!
de_dust, CS :D nostalgic
Shoot, I was late 😂
Good old times
@@sebastiankamph Yeah last time I played was when we switched from 56K modem to ADSL. I was really bad at it, couldn't blame ping anymore 😄
Such a good map!
@@kiksu1 God, what an amazing era, when bunny hopping was barely known.
Nice! Where can you get the Supir image upscaler? I would like to try it out.
The guy didn't even use it, he just did the cherry picked examples on their website. This shit is as bad as 2MP.
How do you think tile multi-diffusion upscaling in A1111/forge compares with Suir? Obviously the resource cost is much higher, I wouldn't expect it to be equal.
I think the Supir upscaler now works on 12GB vram cards. But I couldn't get it to download.
I'm very excited about that new AI restore tool. I mainly use AI to do restorations, so it might be quite useful.
It does look very promising indeed!
Is supir already available? The link leads to a demo…. so whats the point?
It is available, you can check it on Github for example for some more info on how to get it running.
@@sebastiankamph ok…🙏🏻
Google changed your prompts without asking you. Just think about it for a minute ...
it’s a scary thought
I watch your videos for the intro whisper: AI!! well, and for the info too 😂
Legend! Maybe I need to have some asmr whispers as Patreon content?
@@sebastiankamph woah!!! all the dad jokes in whisper mode... LOL
I am running Supir in comfyui with a RTX 3060 12gb and its pretty insane, the best upscale/enhance that i ever tested!
Oh, it's already available at that level? That's pretty wild
@@sebastiankamph there already some threads in reddit about it, its working nicely. The upscale of my dreams :))))
@@sebastiankamph yes already available for comfyui! The best upscale for sure, cant compare to any other available!
Make everybody stupid to the point of learning programming languages becomes an obsolete skill. Nobody will have to think to solve any kind of problem.
I'm all for AI, 100%, big user. But EVERYBODY needs to continue or start learning how to do things with their own brain.
Lot of promises and marketing from tech companies. We are not there yet.
Oh yeah, Dust 2 was one of my favorite maps in that old game :P
I heard they updated and revised it some years later
Dust 2 and Militia were staples.
3:07 Jensen Huang stole my old thoughts 🙂
tutorial of supir installation please !!!!!
counterstrike (the image you asked about?)
5:15 it's for sure tetris!
A manufacturer of bikes might say, no one needs to walk anymore, but they woun't, as everyone knows walking is good for the body.
Similarly kids today unfortunately are NOT getting smarter then the previous generation, the illusion is present because they have a waaaaay easier access to information like search engines and AI, but as soon as that is taken away in any situation the truth leaks out. It's just to make people more and mroe dependent on the product.
so when can i use SUPIR?
Right now
@@sebastiankamph how?! 🤔
Pretty sure that game screenshot is Counterstrike
Hey Sebastian, what happened to your free A1111 styles? Are they not free anymore?
They are now behind the $2 paywall, where you also get all my guides :)
I am actually grateful to Google for letting me know so clearly that they are ready and willing to elevate political correctness above accuracy until it is so egregious that the public basically riots. Now I don't have to guess about it and I can safely withhold giving them a dime in premium subscription fees. Thanks, Google!
It would seem we are kindred spirits.
They knew all along what they were coding. F them to the bitter end!
The next versions will just be more sneaky about it.
@@realWorsin Yea boi...
It looks to me they were trying to solve a real problem of most ai text to image generators
, but went too far trying and overcorrect it. I don't know why people get so pissy about it, they literally took it down and are trying to fix it.
supir looks nice but it's hard to tell how much of that is true when the examples are just two superimposed images switching with javascript/CSS code and not the real workflow.
when i see the performance of SORA, i can belive that videogames could be created by AI in the future... far furture
why far?
my childhood game!
I really hope I'm not so old that de_dust is really a point of trivia now. 😢
Excellent content, but please leave out the background "music", it's really irritating.
I found that lots of models actually skew Asian as well, it's all about the trained data. But I've never seen a model just give me the wrong thing. If I ask for a black person I get one. Not really sure what the issues was. Bard is just broke, just as much as google is.
I think there are some good non-political reasons for wanting to switch up the ethnicities of the characters generated. Like in the old "This Person Does Not Exist" GAN that's always been the case. But now if tell the image generator to, say, create a bunch of birthday partiers, it's always just a bunch of white kids. Just for realism's sake there should be a bunch of black or Asian-looking people among them, without me having to specifically mention that in the prompt. So it makes sense to manipulate the algorithm that way. But if you do that without taking into account that people in an historical context tend to look a certain way you get stuff like that.
Let the AI be truthful and if you want it to lie then use your prompt to do it.
@@fredmcveigh9877 When asked to draw a man on the street a model always drawing a middle-aged white man downtown is not the same as "the truth." Some models are just insufficient and badly weighted. So how do you get it to sometimes create the suburbs or a village and sometimes a person of color and of varying age?
Like I said, there aren't just political reasons for avoiding the model falling into too simple a groove.
If you want a black, blue, green or whatever male then prompt for it. It is being petty and extreme to expect at this moment in AI history, political correctness. Given time, like in real life things will be sorted out.@@OolTube02
If i use an AI tool to code, i need my coding skills to correct what i wasnt able to express correctly in human language. If i use an AI to translate something, i need my language skills to correct what the AI didnt get of what ive told him. To make our language "AI compatible" we have to make it non polysemic (or use a Lojban style language) and also we have to manage incompleteness whereever it is. In someway, WE will have to think like machines if we want to harness the full power of AI. Meanwhile, i ll do some coding tutorials...
This should not have made it past internal testing.
Supir VRAM requires 30 GB X 2. this is insane
No AI gonna make spaghetti code like I do ...
Yes they can. Ask ChatGPT to write a code in gw-basic.
i am still on waiting list for 1.5 gemini. been a few weeks now, i hope to be able to feed it soon
What will you try first?
@@sebastiankamphI have no idea really. Haha. I’ll probably get it to analyze a bunch a data, maybe test its coding ability. I don’t know yet. Haha.
Counter Strike :)
It is fun how we try to adjust the AI rather than talk about the main problem that is humans.
AI is a strict logical sorting routine that shall simulate neutrons, but the data it is trained on is from Internet and there is where we find the main problem.
We should not try correct the AI that do nothing wrong, it is only when humans try to influence it we run into problems. If we ask the AI something and it say "Hi77ler did nothing wrong", then that is a problem with Twitter or where it got the information from, it is a pure human problem highlighted by the AI and we need to get rid of stupid and hateful on the internet.
The game looks like S.O.F :)
👋
You said Supir requires 30G of VRAM; does that mean you didn't try to run on your RTX4090? Thanks.
EDIT: Never mind. I just read this.. wow.
"Due to the large RAM (60G) and VRAM (30G x2) costs of SUPIR, we are working on the online demo releasing."
👍👍👍👍
Heh, Counter Strike ;)
Too easy? ;)
Coding with AI, we're pretty much already there. We already have tools like Sourcegraph Cody that integrate AI straight into VS code and can effectively function as the translator between human and code, even generate entire scripts provided you have a basic understanding of what kind of things you're looking for. For all intents and purposes _writing_ code is already practically obsolete.
"provided you have a basic understanding of what kind of things you're looking for", If you don't know any code then how can you say that?
@@fredmcveigh9877 Well it certainly gets exponentially faster if you understand _some_ programming logic. If you know the steps to get from A to D, you can directly start asking for B and C. If you know precisely nothing, you can get started by describing your desired functionality and letting the AI figure out the steps to get there and start working through them one by one. If you can't follow the code it's suggesting you, you can ask it to be explained, line by line if need be. That makes it a great learning tool too, it's like having your own pocket tutor.
If you know _some_ code, AI can make writing it extremely fast. You need a method that does x y z, you can poop out a functioning block of code just by describing it. If you run into an edge case neither of you took into account, you can usually get a solution to that as well once identified. If you run into a problem you can't identify, odds are AI might have some ideas. Crazy times we living in.
DE_DUST!
Played a lot? :D
de_dust
Yes, prompt engineering is the new must-have skill. We won't be coding, we will be training.
Prompt engineering. That's a high falutin' term for wording a prompt correctly.
This will revolutionize the animated porn category.
Lost control.... of people's minds... they mean.
Diversity = Money
not for long
@@anaoicar true!
Looks like textures for games can now be using upscaler, so the size of the game folder will finally go below 100gb. But it should be running locally in order to be usable in production.
I'd rather just store high res textures than calculate them on my machine!
@@tbird81 yeah, this brings the problem of forever growing size of the game and endless loading screens. Maybe scale up streaming solution will work perfectly. Well, I may be wrong. Everything needs a god damn testing, you know.
Woke is a joke. Fight it with all your possible forses!
Sorry what are you saying at 5:02
subtitles aren't helping as it says run pod
EDIT:
Lmao it's actualy called runpod
Never mind. Thanks for the video lol
Being 'too diverse' is absolutely a thing. You're an absolute coward for even questioning that.
I have to commend Google on admitting fault with Gemini. I can give them a pass that the images were ugly as Hell stylistically, but the AI response as to why it did what it did is hilariously, and completely, racist. I can't say offensive, because it's such a joke, but it's really funny how bad it was coming from such a top notch company.
They didn't admit fault. They never apologised.
And never admitted the problem was they were surpressing images of white people.
The word you are looking for is RACIST not diverse
Dont become a programmer, dont become anything... because by the time programmers are out of a job so is the rest of the world. You will be told what to do and you will be happy.
Imagine being offended by an image generator
Depends on the image generated.
LMAO
@@fredmcveigh9877 Unless it's deepfake hentais you should be okay
CS de_dust2, not de_dust, that's a different map. No one else got it right in the comments? I win? 🤔
😂😂 you’re right! it’s been a long time…