Backprop has been the only credit assignment algorithm capable... of training large neural nets. Introducing Dust, the first zeroth-order method to pretrain transformers to approach and sometimes even *exceed* backprop with large amounts of computation (bitter lesson!) - Dust is on the order of 1,000 to 10,000x more compute efficient than EGGROLL, the state-of-the-art ES method, for training transformers. - We chose pretraining transformers because it's arguably the hardest task possible for zeroth-order optimization. - Search in high dimensions is really misunderstood: zeroth-order usually gets better with overparameterization, not worse, i.e. at a fixed population size, larger models (up to 120x) reach a lower loss than smaller ones. - Backprop-like gradients emerge from a large population of activation perturbations, and the alignment holds up at every scale we tested, up to 1B tokens, which is critical for scaling. The core idea is a *virtual population*, which helps scale to large populations in transformers by perturbing activations in parallel. w/ Bishwas, serdarml, Akshayshow more

Ukrainian ballistic missile “Pelican” (FP-7) was used for the... first time, with a range of up to 3000 km.. The impact was only 15 km from the Kremlin — that is exactly the distance at which the Moscow oil refinery is located. There is a large-scale fire there.show more

🚨 Luton has more than 26 registered mosques and... Islamic community spaces Luton Central Mosque is The largest and first purpose-built mosque in Bedfordshire, holding a capacity of up to 4,500 worshippers. Muslim population is 74,191, which makes up 32.9% of the town's total population.show more

⚡️BREAKING First large-scale protest in the city of Abdanan... This protest is taking place without riots Abdanan has a population of around 35,000 and is located near Iraqshow more

MaskINT: Video Editing via Interpolative Non-autoregressive Masked Transformers paper... page: Recent advances in generative AI have significantly enhanced image and video editing, particularly in the context of text prompt control. State-of-the-art approaches predominantly rely on diffusion models to accomplish these tasks. However, the computational demands of diffusion-based methods are substantial, often necessitating large-scale paired datasets for training, and therefore challenging the deployment in practical applications. This study addresses this challenge by breaking down the text-based video editing process into two separate stages. In the first stage, we leverage an existing text-to-image diffusion model to simultaneously edit a few keyframes without additional fine-tuning. In the second stage, we introduce an efficient model called MaskINT, which is built on non-autoregressive masked generative transformers and specializes in frame interpolation between the keyframes, benefiting from structural guidance provided by intermediate frames. Our comprehensive set of experiments illustrates the efficacy and efficiency of MaskINT when compared to other diffusion-based methodologies. This research offers a practical solution for text-based video editing and showcases the potential of non-autoregressive masked generative transformers in this domain.show more

🤧 A hefty plume of Saharan Dust will likely... push into Florida by the weekend… This would lead to lower air quality and possible irritation of respiratory conditions. On the bright side, the dust is currently playing a large role in inhibiting tropical activity!show more

TEST FOOTAGE: We are pushing the limits of what's... possible. Exploring new approaches to make the visuals better. We are always trying to outperform our last work. And this is the next big leap. Transformers Arrival in Unreal Engine 5! #Transformers #TransformersArrival.show more

A large-scale fire has destroyed more than 20 buses... at a garage belonging to the Anchieta company in the Dom Cabral neighbourhood of Belo Horizonte in Brazil.show more

More from Belgorod, which is subject to large-scale missile... and GMLRS attacks at the moment. #Russiashow more

The large-scale jihadi advance toward #Souaida — reminiscent of... a scene from The Lord of the Rings — reportedly includes both #ISIS fighters and Ikhwan cadres, according to inside sources. Everything is out for the world to see, and for the American public to judge. Who are we with?show more

😳 A Foreigner Shocked by the Huge Variety of... Products in Ukrainian Stores By the way, this is one of the reasons why some are afraid to accept Ukraine into the EU, because our products are of higher quality and lower price. Ukraine’s accession to the EU would mean that products in stores would have to become better in quality and cheaper in order to compete with Ukraine. For ordinary people this is good, but companies and farmers would lose part of their income. That is why some believe it is better not to accept Ukraine into the EU than to improve the quality of life and health of their own population.show more

The rapid evolution of Russian drones continues, large scale... serial production of "Geran4" and "Geran5," is expected, Ukraines already fragile energy grid will become the prime target ahead of Winter. The updated "G4" can reach speeds of up to 500 km/h and carries a larger payload The pictured "Geran-5" has been transformed into a type of cruise missile with a heavy 90 kg warhead and a range of up to 1000 km.show more

Why not envision a church in Tamil Nadu that... can host over 25,000 worshippers in a single service? With Christians making up more than 15% of the state’s population — outnumbering even the Muslim community — the demand is undeniable. And with a sprawling 25‑acre property in Chennai, the scale isn’t just possible, it’s inevitable and many more to come in the near future . We do not require foriegn funds ,people of India keep giving funds has gifts . Enjoy the service below 👇show more

Israel’s Iron Dome is beginning to show signs of... strain. 🇮🇱🇮🇷 After three weeks of war, the barrage of missiles has led to intensive use of interceptors, and reserves are beginning to dwindle dangerously, which has caused the overall success rate, which had been high until then, to decline. As a result, the number of Iranian missile strikes is increasing, causing extensive material damage and human casualties (the latter being lower than the former due to the high use of shelters by Israeli citizens, which has meant that the death toll is much lower compared to Iran). With a reduction in the interception rate and an ever-escalating conflict, there is concern in Israel that it will not be able to prevent attacks in the near future on ‘sensitive’ infrastructure, which would deal a severe blow to the energy and vital capabilities of the state and its population.show more

Something wicked this way comes for the season: Alisa's... physical on Nintendo Switch is now available to order, at a variety of stockists linked below. This isn't the only Halloween surprise we have in order for you, by the way...show more

We just ran one of the largest robot models... that exists today, all without a local GPU. And it’s running real-time at <50 ms overhead. Scaling laws are becoming real for physical AI. Perceptron’s Isaac 0.5 is one of the largest open-source models, sitting at 36B total parameters. Perceptron AI joins other frontier robotics labs like Dyna Robotics and Generalist in showing the need to massively scale pre-training data, from a 1000s to millions of hours of video. Specifically, they show that the more egocentric video the model is trained on, the less teleop data that it needs to achieve the same loss. At Dreamscale Labs (YC F26) (YC F26), we’ve noticed an important truth here: data and training compute do not just scale in isolation. The model sizes must increase too, and GPT-6 Astra has given a glimpse into what a trillion-parameter robotics model could do. We’re building to serve that future - one that is not constrained by the bottlenecks of on-board compute, but rather, allowing large models to run fast over the cloud. Reach out if you’re interested in our work.show more

70% of AI compute in 2026 is inference. Not... training. The models are built. The real infrastructure challenge is serving them at scale to millions of users and agents, every day, in production. Most enterprise infrastructure is still optimized for the wrong workload.🧵show more