Craig Hepburn

It’s only been a few hours and already my feed is overflowing with examples of Luma AI latest text or image to video solution that might give Open AI a hurry up. This example uses an image created in MidJourney then animated using Luma’s new Dream Machine. What you can then do is enhance the quality of the footage using krea.ai video enhancer. Crazy where things are headed. #genAIvideo #lumalabs #videoAI

34

23 Kommentare

So many people have shared this piece already but - yeah! - it's a great read. It's hard for 99.9% of us to offer anything close to a joined-up opinion about generative AI. As much as there are parts about it we may find grim and soulless, it seems a little churlish to completely write something off that, in ways outside of our comprehension (well, mine, at least), might be just about to solve global famine or cure cancer. But staying in my lane, and looking at what I see in the industries and sectors closest to me, I largely see establishment figures doing establishment things with it. There’s a dull rationality to it all, brought to ‘life’ through bad terminology, awful imagery and, let’s be honest, plenty of chancers who hope to (but mainly won’t) strike it rich. An obvious comparison to make is with the emergence of the internet. I recall feeling much more positive about that. The fact I was 25 years younger might have something to do with it, but I think it goes further than just being older now and suspicious of ‘new things’. Wherever it's ended up, the internet started from a better place. It had roots in sub-culture and counter culture (okay, also in the military). DIY homebrewers and hackers. Extensions on the Whole Earth Catalog principles. Communities and activism. Cyberpunks and aspirations of an open and free platform for information sharing. It inspired brave and different startups - including agencies that did amazingly creative digital things. There are exceptions, of course (and for sure some interesting tools to use), but by and large generative AI has started out less open source and with far more corporate dominance. Moreover, I get the impression that if today you floated an idea for a business that didn’t involve AI in a muscle flexing, ‘powered by…’ fashion, and were looking for some support (possibly financial, maybe even just moral), you’d get laughed out of the room. Are we really that lacking in imagination? Is this truly the kind of monochrome existence we all want to be a part of? I can think of few things that would be more refreshing at present than some startup ideas that do not lead with a loud trumpeting of AI. These feel like just the type of endeavours many of us would really love to hear about. Not so long ago, I had vague hopes that, after near quarter of a century of tech bro dominance, we might be at the advent of a new, fresh, different phase of entrepreneurialism (never liked that word much; always loved the possibilities offered by creating things from scratch). Even if it is hard to see a way to that right now, I'll stay low-key positive it'll surface - some day! - as a counter to the uninspiring current state of play.

133

25 Kommentare

#AI is revolutionizing the music industry, particularly in music video production. AI-enhanced videos deepen emotional connections between artists and audiences by creating visually stunning content that adapts in real-time to viewer reactions and preferences. AI analyzes music data, such as rhythm and lyrics, to generate complementary visuals, creating a more immersive experience. Additionally, AI can personalize videos to individual tastes and emotions, fostering a unique and personal viewer experience. This technology allows artists more creative freedom and collaboration, transforming how we experience music and connect with artists #2immersive4u #AIMusicVideos

3

#ELI5! OpenAI used a game to help AI models explain themselves better! "Prover-Verifier Games Improve Legibility of LLM Outputs." The game pairs two AI models together — a more powerful and intelligent “prover” and a less powerful “verifier” and asks them to essentially outwit one another. The prover’s goal is to always get the verifier to believe in a certain answer regardless of whether or not it is the correct one, while the verifier’s goal is to always select the correct answer no matter what the prover may say or try to persuade otherwise. #Prover #AI #Verifier #LLM #Legibility #OpenAI

24

https://lnkd.in/ecQ3MS9N Interesting demo Gen AI from LumaLabsAI. Note the presumptions of ageing though. AI-generated ageing videos may not accurately represent how a person will actually age, potentially creating unrealistic expectations or anxieties. I get it's a demo, but interested in the ethical side of demonstrating it for this purpose? I think it's different from medical mapping of potential illnesses.

3

Interesting. Bloomberg reports, via sources, that no money is exchanging hands between Apple and OpenAI for usage of ChatGPT for iPhone users within its Siri product as Apple feels that the exposure to its over 1.4B iPhone users is payment enough. So, on the one hand, Apple is utilizing ChatGPT functionality to assist in converting iPhone users into phone upgrades and on the other hand, ChatGPT becomes more of a brand AI name than it is already. The only challenge I have with this is the fact that ChatGPT will be incurring additional cloud-computing costs for usage of its product without the same brand exposure of a user tapping in via its app or desktop platform. Also, since Apple has noted that it would integrate other AI-based products as well into the iPhone, there is no exclusivity either. Now, this could certainly change for the future and the exposure to additional potential paying subscribers is good for the marketing funnel, but I'm just not so sure how good this particular exposure is in converting subs to paid as they won't really be exposed to the differentiation incentivizing them to pay. #apple #ai #chatgpt #openai

18

9 Kommentare

#AI Super Resolution. For those who saw my #CTVWS and CTAM Europe talks, I spoke about using AI in video super resolution. This can bring new life to library standard definition content by making it high definition for #FAST for instance. Bitmovin is one of the companies working in this space. Andy Francis explains we can now do this using a part of #machinelearning called Deep Neural Networks (DNNs). Bitmovin's ATHENA research lab partner proposes DeepStream--scalable content-aware per-title encoding, which provides higher video quality on GPU-based end-user devices. When combined with DeepCABAC, context-aware binary arithmetic encoding for DNN compression, shows ~35% bitrate savings while maintaining perceptual quality (PSNR and VMAF). (I dug into the paper for a bit more understanding.) They're also partnering with Pixop who do AI upscaling.

16

1 Kommentar

Will we soon see long-form fiction entirely created by AI? The development of AI for video creation opens up incredible possibilities. However, one main challenge remains: the lack of control in generating images and scenes. Open-source tools, while offering better control, are still limited in terms of visual quality and animation. On the other hand, proprietary tools focus on hyper-realistic renders, with increasingly fluid animation and higher-quality images, although improvements in codecs and color depth are still needed. But that’s not enough. The consistency of characters and environments is also improving, but essential control is still lacking in areas such as:  - Framing and lighting  - Character expressions and acting - Movement and interaction of characters in complex environments  - Shot reverse angles and scene transitions  - The difficulty of animating multiple characters in complex dynamic scenes For entire fictions to be generated by AI, these issues must be resolved. Eventually, it won't just be a single director leading a film, but a hybrid team of traditional skills and AI experts covering all stages of creation:  - Casting  - Framing and lighting  - Costumes and set design  - Acting  - Editing and color grading The future of audiovisual creation is undergoing a major transformation, and those who master these hybrid tools will have unprecedented artistic and economic control. How can we achieve this mastery?

11

1 Kommentar

🚨 Fascinating AI paper alert: "Consent and Compensation: Resolving Generative AI’s Copyright Crisis" by Frank Pasquale & Haochen Sun is a must-read for everyone interested in AI, copyright, and artists' rights. Quotes: "The opacity and scale of AI systems is disrupting the knowledge ecosystem by significantly eroding authors’ proprietary control of their works, well beyond extant digital practices that have already undermined many authors’ well-being. Whereas prior scraping at scale tended to be focused on the non-expressive aspects of works (such as facts), AI is focused by many prompts on their expressive dimensions. Search engines have historically provided links which lead users to works themselves. In contrast, AI tends to provide substitutes for such works, while failing to provide citations to the works in the dataset most similar to the texts, images, and videos it presents as a computed synthesis." (pages 8-9) - "Under the proposed mechanism, copyright owners can first request AI providers to take actions to effectively prevent their systems from generating outputs that appear identical or substantially similar to relevant copyrighted works. A copyright owner would be entitled to send a notice to an AI provider when he or she identifies that an output generated by the provider’s AI system contains either a verbatim or substantially similar copy of his or her work, or a derivative work. In the notice, the copyright owner would be obliged to document the unauthorized reproduction of the work and his or her copyright ownership, along with a digital copy or an online link to the work." (page 21) - "Given the complexity of the AI supply chain, particularly with respect to generative AI, it is not feasible to impose a per-device cost on AI providers. However, other triggers for payment are possible. Levies on the use of particular datasets may be imposed, or on model training, or on some aggregate number of responses provided to users, or on paid subscriptions. Alternatively, the level of the levy could be benchmarked with respect to some percentage of AI providers’ expenditures or revenues" (page 39) ➡ Link to the paper below. #AI #copyright #consent #AIregulation #AIpolicy #AItraining

170

25 Kommentare

Just used Luma AI (a video creation tool like SORA) on a basic video prompt: 'a black cat walking through an overgrown garden with a brick house in the background.' Let's face it, this is quite good but vs stock library footage for something so simple it's actually inferior. Mind you, my prompt was basic and I need to do some more animation based styles.

4

7 Kommentare

When a loved one dies… is it ok to turn them into an ‘immortal’ avatar? This morning, MIT published a fascinating report on a new booming market in China… Deep fakes of your dead loved ones. Sold as ‘clones’, AI companies are now offering the ability to preserve, animate, and interact with lost loved ones. You can even video call them. I’ve written a lot about how AI is increasingly being used ‘immortalise’ great minds and famous personalities (for anyone interested, I’ve included lots of examples in the comments below, from Steve Jobs to Darth Vader). … but this feels different. The argument is that these clones help relatives better process grief and keep memories of loved ones ‘alive’. Interestingly, this builds on a long cultural history in China of communicating with the dead. So… Just an old approach with new technology? Or a step too far? One founder argues: “In its essence, it’s not much different from a traditional portrait, except that it’s interactive” Unsurprisingly, many disagree. “People may end up in staying longer in mourning.” I’m in no way qualified to comment on the psychology of grief… but reading today’s report on this new ‘digital afterlife industry’, an old adage comes to mind: Just because you can, doesn’t mean you should. A saying that seems increasingly relevant in the age of AI.

451

117 Kommentare

The battle here is over whether OpenAI overstepped legal bounds with a voice for its chatbot (Sky) which sounds eerily close to the actress Scarlett Johansson, especially as OpenAI's CEO Sam Altman posted on X referring to 'Her' https://lnkd.in/e5fxj24t, the movie in which Johansson's voice was used as the voice of the machine. Also, Altman had been courting the actress for months to use her voice, even though she declined ultimately. Now, I went to look for a vocal comparison and came across this one. ⭐️ Vocal comparison between OpenAI's Sky and Scarlett Johansson https://lnkd.in/e9aPqjGa Upon a listen, the voices are similar though you can hear the difference. Even though they are somewhat different, if OpenAI benefited from an imitation of her voice to deceive the public for promotional gains, she could be subject to legal recourse. Also, while voices aren't copyrighted, some recordings of them are. In short, if the voice used seems to confuse the public into believing it was her when it wasn't, there seems to be a stronger legal challenge and in my humble opinion, that's exactly what happened here given the promotion by Altman, his seeking her out, and ultimately the voice which sounds very close even though the actress declined. For the record, I have no legal background whatsoever so grain of 🧂 on any legal opinion from me. That said, OpenAI pausing use of the Sky voice does seem to admit some suspicious behavior. We shall see... #ai #openai

13

15 Kommentare

3-minute read (thank you SwissCognitive for sharing): In the latest example of coopetition, Microsoft, Meta, Intel, AMD, Google and others have formed a new group to design next-gen interconnects for AI accelerator hardware. #ai #innovation #coopetition

3

Just as Legos by themselves represent only a fraction of the complete model, they are literally the building blocks or interconnected elements which when aligned form the whole. Perhaps ultra compact LLM’s might combine as needed like elements which form compounds - when chemicals combine, it’s knowing their properties before which combinations work. Or perhaps this is just the product of an active imagination. <g> “One of the most promising areas of AI development is neurosymbolic AI. This hybrid approach combines the pattern recognition capabilities of neural networks with the logical reasoning of symbolic AI.” #ai #llms

1

This version of CSGO has NO GAME ENGINE. 😱 Yes, you read that right. Welcome to Generative AI Gaming- YEAR 1 😲 This PLAYABLE demo was created instead with a diffusion for world modeling based on Counter-Strike: Global Offensive (CS:GO) Looking forward in how this era of gaming will dramatically change by end of 2025. Would you play a Generative-AI game? Where do you see the future of gaming heading? The link to this demo is in the comments

2

2 Kommentare