Video Creation with VideoPoet's LLM

Today:

VideoPoet: A large language model for zero-shot video generation

Google engineers Dan Kondratyuk and David Ross introduced VideoPoet, a cutting-edge language model designed for making videos from scratch.It's great at creating big, smooth movements without glitches. It can turn text, images, or videos into new videos, style them, fix or add parts, and even match them with sound. 

Unlike most video makers that use diffusion methods, VideoPoet uses language model skills, known for handling different stuff like words, code, and sounds really well. It learns from many types of data and turns them into video, image, or sound tokens.

VideoPoet's cool because it can make longer videos by building on the last bit of footage. It's good at keeping things looking right over time. It also lets people tweak videos it's made, changing what's happening in them. Plus, it can add motion to still pictures based on what you tell it. VideoPoet stands out in tests, with folks often liking its videos more for sticking to the script and being more interesting.

READ THE ARTICLE ON GOOGLE

Microsoft Copilot gets a music creation feature via Suno integration

Microsoft's chatbot Copilot is now a budding music producer, teaming up with the music app Suno. Type something like "make me a pop song about family fun," and Suno whips up a whole track—words, tunes, even vocals. To get this going, open Microsoft Edge, head to Copilot's website, log in, and hit up the Suno plug-in.

Artists and AI aren't always on the same page, especially when AI learns from their music without asking or paying. Even the Grammys are saying no to full AI songs. Suno's a bit hush-hush about where it gets its AI training tunes and doesn't sweat it if you ask for a song "in the style of" someone famous.

While the legal stuff is still a big question mark, homemade AI tracks are hitting it big online, even as music labels push back. But AI music makers are just finding new spots to hang out. And hey, there's talk in the Senate about giving artists more control over their digital look and sound. 

READ THE ARTICLE ON MICROSOFT

Anthropic will help users if they get sued for copyright infringement

Anthropic stepping up big time for folks using their AI chatbot, Claude. Starting January 1st, 2024, they're gonna have your back if you get hit with a copyright lawsuit. They're not just gonna defend you in court, but they'll also cover any settlements. 

This move's in line with what other big tech companies like Microsoft and Google are doing. But here's the catch: this protection won't cover you if you're knowingly ripping off someone else's work. Anthropic's been clear about this, even telling the US Copyright Office that they don't think it's cool to use Claude to copy copyrighted stuff.

READ THE ARTICLE ON THE VERGE

Harvey raises $80M Series B from Elad Gil, Kleiner Perkins, OpenAI and Sequoia

Harvey just bagged a cool $80 million in a Series B funding round, bumping up their total cash to over $100 million and rocketing their value to $715 million. Big names like Elad Gil and Kleiner Perkins led the charge, with others like OpenAI Startup Fund and Sequoia pitching in. Harvey's thanking their old buddies Elad, Pat, Charlie, Brad, and Ian for sticking around, and they're stoked to work with new pals Ilya and Mamoon from Kleiner Perkins. 

Harvey's big on using fancy AI to make tough, important jobs easier and they've become a go-to for top law firms and financial bigwigs. Their revenue's shot up big time this past year and they're pulling together a killer team. With this new cash, they're planning to beef up their AI game, grow their team, and add some cool new features. They're all about pushing the limits of AI in professional services and making sure their customers get the best out of it.

READ THE ARTICLE ON HARVEY

Google Funds Leading Civil Rights Group’s New AI Policy Center

Google has funded a new center dedicated to developing policies for artificial intelligence (AI) at a prominent civil rights organization. This move is part of Google's broader effort to shape AI development in a manner that advances equity and addresses potential biases, especially those affecting communities of color. The center is aimed at ensuring that racial equity policies are integrated into AI technology.

This initiative by Google, through its philanthropic arm, Google.org, was part of a larger $20 million grant for AI policy initiatives. The center, named the "Center for Civil Rights and Technology," was launched by the Leadership Conference on Civil and Human Rights, a coalition of over 230 groups. It was introduced as a first-of-its-kind research and advocacy center focused on the intersection of technology, AI, and civil rights.

The Leadership Conference's new AI policy center plans to focus on advocating for legislation and regulation that protects communities of color from potential AI biases and harms. This includes addressing issues like facial recognition technology that misidentifies darker-skinned individuals and AI systems that may perpetuate racial stereotypes or discrimination in hiring practices. Despite Google's funding, the center asserts its independence, with a commitment to advancing technology that promotes, rather than reverses, equity.

READ THE ARTICLE ON ONE NEWS PAGE

Bluestone Equity Partners Has Taken $45 Million Stake In An AI-Driven Video Editing Firm That Helps Connect Fans With Live Sports

Bluestone Equity Partners, led by ex-NBA bigwig Bobby Sharma, just made a major move by dropping $45 million on VideoVerse, an AI-powered video-editing company. VideoVerse is all about making snappy video highlights for social media, especially from live sports. They're big in the game, working with top sports events like the UEFA Champions League and Wimbledon. 

This investment is Bluestone's largest yet and marks their entry into the software-as-a-service space. Despite the tough economic climate, live sports are still attracting big money. Bluestone's betting big on VideoVerse to grow their reach and jump into new markets.

READ THE ARTICLE ON BUSINESS INSIDER

A new supercomputer aims to closely mimic the human brain — it could help unlock the secrets of the mind and advance AI

Aussie researchers are launching a supercomputer, DeepSouth, in April 2024, that's as brainy as our noggin, doing 228 trillion things a second. It's the first of its kind, simulating brainy bits like neurons and synapses big time.

DeepSouth is all about neuromorphic computing, copying how our brains work. It's based in Western Sydney University. Our brains are powerhouses, doing crazy amounts of work with less juice than a fridge light. Compare that to the world's top supercomputer, HPE Frontier, which is huge, eats up tons of power, but can't beat our brain's efficiency.

DeepSouth isn't alone. The Human Brain Project in Europe made BrainScaleS, another brainy machine, studying how neurons fire off. These brain-mimicking computers are game-changers, giving us sustainable computing and new ways to get AI smart.

READ THE ARTICLE ON THE CONVERSATION

AI Memory Mirrors Human Brain

Scientists have found out that AI memory works a lot like our brains, especially in the hippocampus, the part of the brain key for making memories stick. They looked at the Transformer model, a big deal in AI, and saw it copies the brain's NMDA receptor, which helps turn short memories into long ones. 

This is a big step for AI and helps us get how our brains handle memories. The Transformer uses a sort of "brain gatekeeper" just like the NMDA receptor does. When they tweaked this in the AI, it remembered stuff better, like boosting brain power in a way. This mix of brain science and AI could lead to smarter, cheaper AI that learns like us and helps us understand our brains better through AI.

READ THE ARTICLE ON NEUROSCIENCE NEWS

Champion for Openness

A bunch of big names like Meta and IBM, plus a mix of other companies, nonprofits, and schools, just kicked off the AI Alliance to push for open source AI. This crew, including tech giants like AMD and Intel, is looking to develop tools and set standards for AI that anyone can use and improve. They're all about making AI safer and more accessible.

But not everyone's on board. Some big players in AI, like Apple and Google, aren't part of this. Also, there's a bit of a tussle over what "open" really means. You've got everything from totally closed-off AI like GPT-4, to stuff that's kind of open but with strings attached, to completely open-source projects.

The big deal here is that more open AI could lead to faster progress and innovation. The AI Alliance is a powerhouse that could really shake things up, standing up to companies that want to keep their AI under lock and key. Open-source fans say letting everyone in on AI makes it safer, since more eyes can spot and fix problems. This Alliance could be a big win for making AI better and more open for everyone.

READ THE ARTICLE ON DEEPLEARNING.AI