Microsoft's Unsettlingly Lifelike AI Avatars

Microsoft has unveiled VASA, an AI framework that generates disturbingly lifelike digital avatars from just one image and an audio clip. The high-fidelity animations capture nuanced facial expressions, lip movements perfectly synced to speech, and natural head motions - making the virtual humans look almost indistinguishable from reality.

Powered by a novel "face latent space" tech trained on videos, VASA can encode the subtle facial cues that sell the illusion of life. The VASA-1 model can render 512x512 videos at up to 40 FPS with negligible lag, enabling real-time digital human engagements.

While intended for positive use cases like tutors or accessibility aids, experts warn VASA could be a powerful deepfake tool. Microsoft acknowledges identifiable artifacts for now but has no release plans until safeguards against misuse are in place.

The benefits are vast - from enhancing education to providing companionship. But the risks of misleading deepfakes are just as concerning. Microsoft opposes tech that deceives by creating misleading content of real people. Yet their own researchers admit a gap remains to achieve true video authenticity.

As the boundaries of AI-generated humans are pushed, the question remains: At what point do these digital avatars become too realistic? For now, the virtual and real remain semi-separate. But that line may soon blur in unsettling ways.

Meta's 'Llama 3' AI Wants a Seat at the Generative AI Table

In the latest AI arms race salvo, Meta just unleashed its new Llama 3 language models - the 8 billion parameter 'Llama 3 8B' and the beefy 70 billion parameter 'Llama 3 70B'. And the social media giant is talking a big game, claiming these llamas are ready to run with the biggest generative AI alpacas on the block.

See, Meta says its new Llama models are knocking it out of the park on all the trendy AI benchmarks the cool kids care about - from coding tests like HumanEval to common sense challenges like BIG-Bench Hard. The 8B version is supposedly outperforming other open 7B models like Mistral and Gemma, while the 70B Llama is aiming to go toe-to-toe with Google's flagship Gemini 1.5 Pro.

Of course, benchmarks will only get you so far in the real world. But Meta seems confident these Llamas, trained on custom 24,000 GPU clusters, have the horsepower to hang with the AI elite. Whether they actually have the smarts to dethrone Gemini and Claude from the generative AI throne remains to be seen. But one thing's for sure - the Llama drama is just getting started.

OpenAI's Global Office Expansion Blitz Continues

The AI upstart that brought you ChatGPT is going big on its real estate footprint - OpenAI is reportedly scoping out spaces in New York City to open its fifth major office location.

After starting 2022 as a humble San Francisco one-office operation, OpenAI has been aggressively expanding its physical presence worldwide at a blistering pace:

• Tokyo office opened to focus on Japanese language models

• India outpost coming with Twitter vet Rishi Jaitly as policy advisor

• Additional SF location being scouted to accommodate 1,000+ employee headcount

• London and Dublin offices established just last year

With over $1 billion in funding from Microsoft and others, Sam Altman's AI juggernaut is leaving no stone unturned in its bid for global domination. An NYC office puts OpenAI in the heart of the east coast tech scene and America's corporate capital. Whether it's Brooklyn or Manhattan, you can bet the rent won't be cheap - but hey, when you're valued at a cool $20 billion, a few million here or there is just a rounding error.

Meet the 'Limitless Pendant' - Your New AI Notetaker

While other AI wearables have tried to replace your smartphone entirely, a new gadget called the Limitless Pendant is taking a different approach - being the smart assistant for all your meetings and calls.

Instead of an all-in-one device, the pendant is designed as more of an accessory to make recalling conversations a breeze. Clip it on your shirt or wear it like a necklace, and the Pendant will use AI to transcribe everything said and even suggest follow-up action items based on the discussion.

But here's the real kicker - the $99 Limitless comes with a free pricing tier that gives you 10 hours of AI features per month along with unlimited audio storage. Want more AI power? The $19/month Pro plan unlocks unlimited transcriptions while still letting you store all recordings for free.

Privacy is also a big focus, with the Pendant requiring consent to record others and keeping all data anonymized and encrypted. And with a 100-hour battery, waterproof aluminum body, and color options from black to hot pink, it checks plenty of boxes.

Of course, the jury's still out on whether this AI meeting companion is truly "limitless" or just another overpriced wearable. But by avoiding trying to be a phone replacement and instead focusing on enhancing actual human interactions, the Limitless Pendant could find its niche.

After the struggles of the Humane AI Pin and uncertainty around the rabbit r1, this privacy-conscious, affordable little accessory might just be the AI wearable we've been waiting for. Or it could be vaporware by August - but hey, at least they're thinking different.