Felipe Meres

Artist & Scholar | Visiting Prof @CooperUnion | PhD in Anthropology Sci&Tech @TheNewSchool | MFA @BardCollege| 3D generalist @Venarie_Studio | He🌈 Him

[Felipe Meres @FelipeMeres]

In this essay I explore a pre-Columbian object's complex mythology and how hyper-real technologies disrupt referentiality & reshape authenticity. It served as a point of departure for my dissertation, completed last year and being prepared for publication. https://t.co/YdSD7THUhA

[Felipe Meres @FelipeMeres]

@franklingraves so as others have pointed out, the ability to extract intact chunks of the training data from the models play a significant role in these discussions

[Felipe Meres @FelipeMeres]

@mmitchell_ai @ravenben If "real damage" equals "affect the model in any significant way", why would it be harder to affect the model for malicious uses than for the intended purpose?

[Felipe Meres @FelipeMeres]

@mmitchell_ai @ravenben I'm confused: "poison samples ... can corrupt an Stable Diffusion SDXL prompt in <100 poison samples" and yet "Zhao admits ... people might abuse the data poisoning technique for malicious uses. However, ... attackers would need thousands ... samples to inflict real damage"

[Felipe Meres @FelipeMeres]

@CriticalAI @DukePress excited to read this tomorrow but for now: thank you for the genius that is the title humanities in the loop

[Felipe Meres @FelipeMeres]

@victorlavalle I just finished the book after starting it after episode 3 of the show. Really fantastic work! I loved the knotty relation between fantasy and technology in your fairy-tale realism - really brilliant use of allegory. A joy to read and think with.

[Felipe Meres @FelipeMeres]

@mmitchell_ai I noticed that as well. Equal rights to vocal fry and upspeak!

[Felipe Meres @FelipeMeres]

@TheBrianPenny @TIME The AI narrative is really bringing this to the forefront. The people at Oxford 🤦🏽‍♂️

[Felipe Meres @FelipeMeres]

@TheBrianPenny @TIME I had to stop at the second paragraph. Wall-e is a much better take.

[Felipe Meres @FelipeMeres]

@TheBrianPenny @TIME The amount of quacks with stable positions at once-prestigious universities is staggering.

[Felipe Meres @FelipeMeres]

@Ola_Michalec 1000 elaborate ways of convincingly saying very little in 20 pages

[Felipe Meres @FelipeMeres]

@Ola_Michalec That sounds like most of the humanities/social sciences to me - you can tell from the first page if the author has something to say or if they are playing the linguistic game

[Felipe Meres @FelipeMeres]

@TheBrianPenny That's a very good point. I think there's a lot to this story that is still under wraps

[Felipe Meres @FelipeMeres]

@TheBrianPenny What has been made public of the deal so far is that Stability offered to cover AWS GPU cost for Bjorn's lab to train the models and in return laid claim to the model, rebranded it from Latent to Stable Diffusion and presented it as it's own creation

[Felipe Meres @FelipeMeres]

@TheBrianPenny Yeah I think you were thinking about the guy who sold his share of the company 7 days before they switched from Covid AI to SD image gen

[Felipe Meres @FelipeMeres]

@TheBrianPenny I was not aware Bjorn had stocks of either Stability or Runway, I would assume him running a lab at an university would make that a pretty clear conflict of interest that would go against university policy? Afaik what motivated him to collab was access to GPUs to train models

[Felipe Meres @FelipeMeres]

@TheBrianPenny (Ommer and his lab were an integral part of developing both Taming Transformers and Latent Diffusion)

[Felipe Meres @FelipeMeres]

@TheBrianPenny I'm sure there are NDAs at play but the little this Forbes article quotes of Björn Ommer suggests that him and his research peers were duped by Mostaque https://t.co/RzSVZEJMQ5

[Felipe Meres @FelipeMeres]

@TheBrianPenny I agree 100%. And I think the researchers/academics who developed the tech for non-commercial purposes would agree as well.

[Felipe Meres @FelipeMeres]

@TheBrianPenny One of the crucial questions is that a lot of these folks — the ones who actually developed the tech — were in fact doing it for research. It was when the midjourneys and stable diffusions sprouted as a response to Dall-e's success that the whole debacle was set in motion

[Felipe Meres @FelipeMeres]

@TheBrianPenny This was one of the notebooks/scripts that spearheaded a lot of the opensource activity around ai image gen: https://t.co/BlbPMqHBOp