Meta Releases 'Human-Like' AI Image Creation Model (reuters.com) 25
Meta said on Tuesday that it would provide researchers with access to components of a new "human-like" artificial intelligence model that it said can analyze and complete unfinished images more accurately than existing models. From a report: The model, I-JEPA, uses background knowledge about the world to fill in missing pieces of images, rather than looking only at nearby pixels like other generative AI models, the company said. That approach incorporates the kind of human-like reasoning advocated by Meta's top AI scientist Yann LeCun and helps the technology to avoid errors that are common to AI-generated images, like hands with extra fingers, it said.
Meta, which owns Facebook and Instagram, is a prolific publisher of open-sourced AI research via its in-house research lab. Chief Executive Mark Zuckerberg has said that sharing models developed by Meta's researchers can help the company by spurring innovation, spotting safety gaps and lowering costs. "For us, it's way better if the industry standardizes on the basic tools that we're using and therefore we can benefit from the improvements that others make," he told investors in April.
Meta, which owns Facebook and Instagram, is a prolific publisher of open-sourced AI research via its in-house research lab. Chief Executive Mark Zuckerberg has said that sharing models developed by Meta's researchers can help the company by spurring innovation, spotting safety gaps and lowering costs. "For us, it's way better if the industry standardizes on the basic tools that we're using and therefore we can benefit from the improvements that others make," he told investors in April.
Does it include legs? (Score:3)
Because Mark's Metaverse had no legs to stand on. /s
Re: (Score:3)
https://github.com/facebookresearch/ijepa [github.com]
Great stuff (Score:3, Insightful)
Meta, rightfully, gets a lot of crap over privacy and censorship issues but there is no denying they've been doing a lot of advance VR/AR/MR and AI technology and that has a real benefit.
Great (Score:2)
and helps the technology to avoid errors that are common to AI-generated images, like hands with extra fingers, it said.
Great, way to piss of the people with Polydactyly.
Ick, eew, yuck (Score:3, Interesting)
That approach incorporates the kind of human-like reasoning....
What a disgusting statement. It is not human like reasoning. It is mathematical pattern matching, and nothing more. That doesn't mean it isn't impressive pattern matching, but stop trying to shoehorn every bit of mathematical prowess into the impossible "human-like" category. There is nothing human-like about it.
Re: (Score:3)
There is however huge marketing potential in claiming "human like" so they are doing it, no matter how big the lie is. In fact, the "Big Lie" approach teaches that truth is not required to sell even expreme ideas or products. You just need to do away with your personal integrity and decency, but many people are willing to do that for a buck.
Re: (Score:3, Interesting)
What a disgusting statement. It is not human like reasoning. It is mathematical pattern matching, and nothing more... There is nothing human-like about it.
I don't know about you, but my human reasoning feels like it's almost entirely pattern-matching. At least so I believe based on introspection, observation on others, my undergrad courses on cognitive neuroscience, and my masters degree in history & philosophy of science. I think that's true in my daily interactions with others, in my professional career as a senior software developer, in how I make sense of politics and literature and the world around me. It makes sense given lab studies which show that
Re: (Score:1)
That is a statement of faith, nothing more.
Re: (Score:2)
Since one can literally change one's pattern matching and get a different result, your assertion merely shows how little you understand about your own mind.
Re: (Score:2)
No, but your assertion shows how much detail your own mind glosses over. The ability to recognize and apply patterns does not make them mathematical, it does guarantee that function can be replicated mathematically, even if it can that does not mean the replication operates in the same way. Nor does it establish this as the totality of the processing occurring in the mind.
Re: (Score:2)
No, but your assertion shows how much detail your own mind glosses over.
How does it?
The ability to recognize and apply patterns does not make them mathematical
I wouldn't describe them as mathematical like the AC did but the whole of physics is mathematical so I could see where he's coming from.
it does guarantee that function can be replicated mathematically, even if it can that does not mean the replication operates in the same way.
One can't even measure the function so we're only replicating a 50s model of neural processing.
Nor does it establish this as the totality of the processing occurring in the mind.
Nobody said it did.
Re: (Score:2)
It looks like I need to reply to you in reverse.
"Nor does it establish this as the totality of the processing occurring in the mind.
Nobody said it did."
Actually they did... everything in this thread is in response to AC doing so:
"> It is mathematical pattern matching, and nothing more.
So is human reasoning, nothing more."
Since we have absolutely no clue how to replicate sentience, I declared this a statement of faith.
"No, but your assertion shows how much detail your own mind glosses over.
How does it?"
Hi
Re: (Score:2)
"Human reasoning" is not "the totality of the processing occurring in the mind". Nor is it sentience.
Reality is not mathematical
It quite possibly is ie simulation theory. Though I don't believe in the latter. Plus "reality" would arguably include the simulator and its designer.
Finally (Score:2)
Meta Releases 'Human-Like' AI Image Creation Model
Something to help Mark find a better haircut. :-)
Re: (Score:2)
Whoa, slow down there, Mark lives comfortably in the uncanny valley. We don't want him crawling out into the human side now, best to keep him where he belongs.
human like image creation (Score:3)
Better models? (Score:2)
Is it better to focus on improving training of models with more data making more of an effort to include clearly visible features (hands, fingers) or bolting on a bunch of additional correcting models to fix what the not good enough original got wrong?
For example: Some people wear sun glasses, others sometimes have them resting on their heads. Assume none of the training data has people wearing sunglasses and a separate pair resting on their heads.
Is it possible to train up a model sufficiently where it do
Game Changer for DeSantis Presidential Campaign (Score:2)
Re: (Score:2)
The Ron DeSantis presidential campaign staff are investigating using this new tool to generate campaign posters of the candidate that are more human-like.
They should contact Dr. Eldon Tyrell who runs a BIG CORP that specializes in products that are MORE HUMAN THAN HUMAN
Wow (Score:1)