Zohran Mamdani as a creepy trick-or-treater, Gavin Newsom body-slamming Donald Trump and Hakeem Jeffries in a sombrero. This is not the setup to an elaborate joke. Instead, these are all examples of recent AI-generated political videos. New easy-to-use tools ' and acceptance of those tools by politicians ' means that these fake videos are quickly becoming commonplace in American politics. Perhaps the most interesting thing about many of the videos is how clearly fake they are. Rather than trying to deceive the viewer into thinking a depicted event actually happened, the videos serve a different purpose. President Trump didn't post a video of himself wearing a crown in a fighter jet dumping feces on a group of protesters because he wanted people to believe that the flight actually happened. He likely did it to express his feelings about the protest and to create an in-joke with his followers. Fears about the political implications of AI-generated videos have been around since the term deepfakes was coined in 2017. Steady improvements in the technology mean that distinguishing real from fake could become a significant threat. But today's use of AI imagery is largely about making memes and making money ' in other words, typical social media content....

Scientists have produced the most detailed 3D map of almost all buildings in the world. The map, called GlobalBuildingAtlas, combines satellite imagery and machine learning to generate 3D models for 97% of buildings on Earth. The data set, published in the open-access journal Earth System Science Data on 1 December1, covers 2.75 billion buildings, each mapped with footprints and heights at a spatial resolution of 3 metres by 3 metres. The 3D map opens new possibilities for disaster risk assessment, climate modelling and urban planning, according to study co-author Xiaoxiang Zhu, an Earth observation data scientist at the Technical University of Munich in Germany. It could also help to improve how researchers monitor United Nations (UN) Sustainable Development Goals for cities and communities, Zhu adds. Conventionally, creating detailed 3D maps at a global scale has been difficult, say Zhu, because it usually requires either laser scanning or high'resolution stereo imagery. The team's solution was to combine deep learning with laser-scanning techniques to predict building heights. The tool was trained on reference data obtained using light detection and ranging (LiDAR) from 168 cities, mostly in Europe, North America and Oceania....

Today, we are going to discuss a highly controversial topic: can we reimagine computers from the ground up for AI' What got me thinking about this idea was the announcement of Unconventional AI which raised a considerable amount of money of work precisesly on this problem. Artificial intelligence today thrives on brute-force computation. Breakthroughs in deep learning, from chatty large language models to expert game-playing agents, have largely come from one approach: throw more matrix-math and more GPUs at the problem. This strategy of scaling models and data on ever-larger GPU clusters has driven impressive gains. However, it's also driving us straight into a wall of diminishing returns ' primarily in the form of exorbitant energy use and hardware costs. Recent events underscore this concern: a new startup called Unconventional AI made headlines by raising an unprecedented $475 million seed round to develop radically new computing hardware for AI. The motivation' Today's AI compute engines are power-hungry and inefficient compared to the elegance of brains. The human brain performs extraordinary feats on only ~20 watts of power, whereas training a single large AI model can devour megawatt-hours. The sheer gap suggests that AI might require a new form of computing to continue its trajectory. In this essay, we'll explore why current AI runs on matrix multiplications and GPUs, why this paradigm is straining, and what 'unconventional' new computing approaches ' from analog chips to photonics, neuromorphic designs, and even quantum systems ' could define the next era....

'There must be only One Rulebook if we are going to continue to lead in AI,' Trump said. 'We are beating ALL COUNTRIES at this point in the race, but that won't last long if we are going to have 50 States, many of them bad actors, involved in RULES and the APPROVAL PROCESS'AI WILL BE DESTROYED IN ITS INFANCY!' Trump's statement comes days after an effort to preempt states from regulating AI was quashed in the Senate, as Congress couldn't agree to insert the deeply unpopular proposal into a must-pass defense budget bill. The fast pace of AI development and the lack of general consumer protections from the federal government has led many states to enact their own rules around the technology. California, for example, has the AI safety and transparency bill SB 53, while Tennessee's ELVIS Act protects musicians and performers from unauthorized AI-generated deepfakes of their voices and likenesses. Silicon Valley figures, including OpenAI President Greg Brockman and VC-turned-White House 'AI czar' David Sacks, have argued that such laws by states would create an unworkable patchwork of laws that would stifle innovation and threaten the U.S.'s lead against China in the race to develop AI technology....