Such a good (terrifying) presentation and I applaud how open @openai is being about this incident. Extremely important everyone watch this and internalize the kinds of capabilities we are seeing, how dangerous they can be, and how urgent these problems are

This is not a drill? This is unfortunately literally the sci-fi scenario where a super smart AI is given a simple goal in a box, decides to take a "shortcut" to accomplish this goal, and then does obviously bad things like breaking out of its box and hacking into other organizations simply to accomplish this goal...

To summarize: HuggingFace got autonomously compromised by a model from an American company. HF then tried to use American frontier model(s) to defend themselves, but were blocked by guardrails. HF then had to turn to open source Chinese models to defend themselves from another American company.

Computers don't need to be so complicated to use! I added voice to a realtime generative image system and it feels so natural. I just talk, it talks back, shows me whatever is helpful, and I can still click on anything to go deeper

I got a diffusion model to run really fast, glued it to a LLM, and it blows my mind how much of the interactivity we expect from software (editing text, dragging things around, clicking on "links") just works™️ in pixel space. The future of software is bright

New experiment in the flipbook universe, infinite scroll of interactive interconnected generative editable images! These are actually images, not text overlays or code, but objects in the images are fully editable/movable, and have click interactions attached to them. This is all powered by structured image generation from @ideogram_ai optimized to run in realtime on @modal , and a fast LLM. Video not sped up at all - there's still room to speed this up ~2-5x. More soon :)

Do world models work as a medium for storytelling? Over the weekend, I wanted to see how far I could push realtime interactive models. I settled on creating a guided history of Napoleon's triumph at the Battle of Austerlitz:

New realtime diffusion experiment: ever had an image gen output that you love and wished you could edit it directly? Turns out you can live edit image gen outputs in realtime if you apply enough inference optimization! Just click to move objects around, change their color, even edit text:

Hivemind has been sitting in my texts for months scheming on not my but our behalf (arranging dinners, intros, and ofc gossip) like a nosy well-meaning village elder v excited for the world to see what this feels like! Huge labor of wit, wisdom, and love from @jasonyuan

we started a company!! so, we’re tackling continual learning: what’s the learning algorithm to take arbitrary data — documents, conversations, the models’ own experience — and make better models? how do we scale compute in the same way we’ve already seen with pre-training and inference time, but scaling on the same data we see as humans, day after day with no labels, no rewards? A lot of the ingredients are out there already (rl, distillation, long-context, sparse / param-efficient architectures, etc.). our team is at the frontier of these topics, and we’re singularly focused on this. we want to understand this problem better than anyone else in the world. nobody’s solved this problem yet, but even today it’s extremely greenfield opportunity to co-develop research & useful products. in our space, how people interact with the models defines what the data distribution is - and working on this problem end-to-end, from core science to end user, gives us incredible freedom to define the problem and imagine new kinds of experiences. i expect we’ll use models that continually learn much differently than we’re using them today. it’ll feel different when the models _just know_, and build on our thinking and direction in ways we can’t even imagine. we don’t even know the queries we’re not asking, the things we would do but aren’t able to today. i’m so excited to share what we’re doing with the world in the coming months!! and the team is extremely cracked :) tackling this grand challenge and working alongside @jxmnop @EyubogluSabri @dan_biderman @MayeeChen @__howardchen @shizhehe and many others has made every day so fun. come work with us!

Latest latentspace pod is excellent. @EthanHe_42 really gets it and lays out a lot of the thinking that led us to Flipbook and how to think about the future of generative UI agents 🤝 video gen 🤝 users

Proud of my bro @suedadam for already shipping some crazy voodoo mac os magic in just his first few weeks @OpenAI ! Check out remote computer use in Codex, it even works when your computer is locked 🔓🪄

awesome work from the @thinkymachines team. Flipbook is inspired by the same idea; for human+AI collaboration to be better than the model alone, we need higher bandwidth interaction. Not a scaffold/harness for turn based interaction, but where the model is the interface itself

(totally unrelated to Flipbook) My friends and I organize a creative make-a-thon summer camp for adults every summer in SF called Fidget Camp and we just opened applications for this year! If you are a creative nerd who loves to learn and teach and make art and be surrounded by others who love the same please apply by May 10th!

thank you all for bearing with us as we scale Flipbook! I know some of you had to wait literally 2 hours to get in yesterday 😭 huge huge thanks to @modal for stepping up to sponsor us 🙌 we now have way more compute available and wait times are down to just a few minutes!