Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Imagen: An AI system that creates photorealistic images from input text

imagen.research.google

Post reply on HN
Print view View original

151–160 of 233 posts

Re: Imagen: An AI system that creates photorealistic images from input text

#151

Post by lm28469 »

IanCal wrote: post #129
lm28469 wrote: post #112

Earlier quoted context omitted.

> I make ai art all day long! I feel like this is the epitome of modern "content creation" Typing a few sentences into a software you barely understand, said software shits 15 jpegs out, 2 are good, "hey I make art". What a sad state of affair, tech is consuming everything and people are cheering, one more step on the path to being complete useless key pressers.

Photography is well known to have killed art. Press a button on a box you barely understand and the camera shits out a pic. "Hey I make art". What a sad state of affair, tech is consuming everything and people are cheering, one more step on the path to being complete useless key pressers.

You still have to decide what you take a pic of, that's the artistic part. You have to chose a subject, lights, &c. it's still incredibly more complex than "hey google draw me a horse with a tuxedo"

You can tell that right away seeing how many camera users never produce anything of quality. A 6 years old can press the button of a camera but no 6 years old will produce meaningful work. In the case of Ai art tools you just need some basic english skills

Re: Imagen: An AI system that creates photorealistic images from input text

#152

Post by lm28469 »

psyc wrote: post #125
lm28469 wrote: post #119

Earlier quoted context omitted.

The user, me included, has less and less knowledge and skills though. When your tool becomes a megacorp owned subscription based service you barely understand is it really a tool ? It doesn't produce art in the way a brush and paint produce art, you barely have any control on what it does, you just become an image filter, you press a few keys, look at the image for 5 seconds and decide if it triggers the right part o…

I can sort of see the "barely understand" part being relevant if I squint. Really don't know how "megacorp subscription" relates to a tool's toolhood.

> Really don't know how "megacorp subscription" relates to a tool's toolhood.

Owning the tool vs being owned by the tool, yadda yadda..

Re: Imagen: An AI system that creates photorealistic images from input text

#153

Post by dumpsterlid »

fvold wrote: post #34
stevewatson301 wrote: post #17

Earlier quoted context omitted.

My guess is they're they'll use Imagen and LaMDA to build a "conversational" search experience of some kind. So, instead of providing a list of websites to go to, they'll synthesize an answer to the search query, with imagery to go along with it, and so on.

I don't know about the market at large, but I do not want that. I want the a search engine to just have a huge database of websites, and look up stuff in that database based on my query, spitting out a link to the page that matches best, then the one that matches second best etc etc. Using ML to determine the order of matches is absolutely fine, but to "digest" the internet and cook up an answer to what I'm looking f…

Yah I would love someone to convince me this isn't like touchscreen interfaces becoming prevalent in cars over physical control surfaces. It is sold as technological advancement that is good for everyone but in reality it is just corporations figuring out how to make something worse for cheaper and force consumers into it.

I already feel like google search sucks. It is partially googles fault partially the internets fault. The best source of information was and will always be forums full of knowledgable people having genuine conversations and everybody moved to discord because.... because... om I dont know and besides google doesnt get paid to elevate tiny ass forum websites in the search results.

Re: Imagen: An AI system that creates photorealistic images from input text

#154

Post by FairlyInvolved »

goldenkey wrote: post #82
FairlyInvolved wrote: post #79

Earlier quoted context omitted.

I kind of get the sentiment about openness but I think it's way more nuanced than you are making out. There are very good reasons for withholding SOTA models, primarily from the info hazard angle and avoiding escalating the capabilities race which is basically the biggest risk we have right now. Google / Deepmind have actually made some good decisions to try and slow down the race (such as waiting to publish).

Capabilities race, seriously? This is not nuclear warfare my guy. It's mathematics.

Nuclear warfare is much less concerning than misaligned AI.

Take a look into scaling laws and alignment concerns, this is a very real challenge and existential risk not some crackpot theory.

Re: Imagen: An AI system that creates photorealistic images from input text

#155

Post by psyc »

lm28469 wrote: post #152
psyc wrote: post #125

Earlier quoted context omitted.

I can sort of see the "barely understand" part being relevant if I squint. Really don't know how "megacorp subscription" relates to a tool's toolhood.

> Really don't know how "megacorp subscription" relates to a tool's toolhood. Owning the tool vs being owned by the tool, yadda yadda..

Oh ok. That makes some sense.

Re: Imagen: An AI system that creates photorealistic images from input text

#156

Post by napolux »

Is anyone aware of people doing the same for NSFW images? They can easily wipe out an entire industry.

No models to pay and to check for legal ages, infinite possibilities: just write the pic you want, the massive body part you want, how many genitals are involved and boom. You have your image.

Re: Imagen: An AI system that creates photorealistic images from input text

#157

Post by motoboi »

gallabytes wrote: post #142
motoboi wrote: post #140

Earlier quoted context omitted.

in the case of imagen, I suppose the cost is at least two orders of magnitude over 500k.

... no it definitely wasn't. that's $50m. read the paper, they tell you how long it took on a v4-256, which you know the public rental price for.

One order of magnitude over 500k it's 5M, two orders is 50M

Re: Imagen: An AI system that creates photorealistic images from input text

#158

Post by masukomi »

gallabytes wrote: post #142
motoboi wrote: post #140

Earlier quoted context omitted.

in the case of imagen, I suppose the cost is at least two orders of magnitude over 500k.

... no it definitely wasn't. that's $50m. read the paper, they tell you how long it took on a v4-256, which you know the public rental price for.

they're talking about the total cost of the project, including the salaries of the humans, their health care, their office space, etc.

Re: Imagen: An AI system that creates photorealistic images from input text

#159

Post by FairlyInvolved »

echelon wrote: post #108
FairlyInvolved wrote: post #79

Earlier quoted context omitted.

I kind of get the sentiment about openness but I think it's way more nuanced than you are making out. There are very good reasons for withholding SOTA models, primarily from the info hazard angle and avoiding escalating the capabilities race which is basically the biggest risk we have right now. Google / Deepmind have actually made some good decisions to try and slow down the race (such as waiting to publish).

They're not slowing down anything. The cat's out of the bag. What good does a few months lag do when nobody is bracing for impact?

I'm not saying they are doing a good enough job, but that doesn't mean their approach isn't entirely without merit.

Even ignoring the infohazard angle if they published everything immediately that would escalate the race. By sitting on their capabilities and waiting for others to publish (e.g. PaLM, Imagen vs GPT-3, DALL-E) they are at least only playing catch up.

Re: Imagen: An AI system that creates photorealistic images from input text

#160

Post by polygamous_bat »

gallabytes wrote: post #142
motoboi wrote: post #140

Earlier quoted context omitted.

in the case of imagen, I suppose the cost is at least two orders of magnitude over 500k.

... no it definitely wasn't. that's $50m. read the paper, they tell you how long it took on a v4-256, which you know the public rental price for.

While I don't agree with gp's price estimate of $50M, you are also forgetting that to train the final model, they had to iterate on the model for research and development. This imposes a significant multiple over the cost to just train the final model.
Post reply on HN