Meta's Move Towards AI Governance

Today:

Meta Ran a Giant Experiment in Governance. Now It’s Turning to AI

Meta ran a pretty grand democratic experiment recently, pulling together around 6,000 folks from across the globe to chat about the company's responsibility in the metaverse they're building. Most folks liked the idea, saying it's a good way for Meta to make decisions. Now, they're looking to run a similar shindig for generative AI, following other tech heavyweights who are also looking into more democratic ways to guide AI.

The process wasn't without its hiccups, though. Some folks felt like they were part of a data collection experiment rather than a democratic process. They had limited ways to interact and didn't get a chance to talk directly with the decision-makers at Meta. Also, Meta's been a bit slow on detailing what actions it'll take based on the feedback they received.

Moving forward, Meta's got to iron out these wrinkles. When they do the same for generative AI, they should look at the best practices from similar government-run processes. The aim is to create a more informed, democratic decision-making process for tech companies, especially when dealing with stuff that affects folks worldwide.

There's no one-size-fits-all answer, but continuing to refine these decision-making processes seems like the best way to navigate the challenges brought on by rapidly advancing AI technology.

READ THE ARTICLE ON WIRED

Netflix’s AI-assisted green screen bathes actors in eye-searing magenta

Netflix has put a spin on the old trick of using a green screen for movie magic, but now they're bathing actors in a vibrant shade of magenta. This quirky method helps to fuse the actor and background more neatly. Normally, actors are positioned against a bright green backdrop, which gets digitally replaced with anything from city skylines to mythical creatures. This method, called "chroma keying," is simple and inexpensive, but it's not perfect. It struggles with see-through objects, fine details like hair, and anything that's similarly colored to the green screen.

Netflix's new method, as explained in a recently published study, involves lighting the actors with a combination of red and blue, resulting in a magenta hue, against a bright green screen. This approach simplifies the task of separating the actor from the background, but makes it tricky to restore the green color to the magenta-lit subjects.

This is where artificial intelligence (AI) comes in. The Netflix team trained a machine learning model to quickly restore the missing green color in a smart way by comparing magenta-lit scenes to scenes lit normally.

There's a catch, though. While this technique can restore colors accurately in post-production, the actors and set have to be lit in a pretty ghastly way. The study suggests flickering the lights really fast to make it appear nearly constant, but this would need some fancy footwork with the camera.

READ THE ARTICLE ON TECHCRUNCH

Urtopia Fusion installs chatGPT in its e-bike to answer riders’ travel questions while cycling

Urtopia Fusion's latest e-bike ain't just any ol' bike, it's got ChatGPT built right in. Unveiled at EUROBIKE 2023, this fancy rig comes with built-in speakers so you can ask the bike travel-related questions and get answers on the fly. Think of it like having your own personal tour guide on the go. The only snag is ChatGPT's last update was in 2021, so it ain't the best for real-time stuff like traffic or weather.

The bike's got other whiz-bang features too, like GPS navigation, anti-theft measures, and safety tech. Plus, it's all voice-activated. They're calling this the first e-bike with a built-in AI chat feature. It's like riding with a know-it-all buddy, only this one doesn't get tired.

And it's not just about looking pretty. The Urtopia Fusion e-bike plays nice with Apple Health and Strava. That means you can keep tabs on your heart rate while you ride and even share your rides on social media. It’s all about making your rides better and getting more folks into cycling.

READ THE ARTICLE ON DESIGNBOOM

Huawei Cloud team says AI model produces faster, more accurate weather forecasts

Chinese tech giant Huawei's researchers claim to have cooked up a super-fast and smart weather-predicting AI, which they've named Pangu-Weather. According to them, this smart tool cranks out forecasts in a heartbeat, detailing all sorts of stuff like temperature, wind speed, humidity, and even a heads up on potential natural disasters.

What sets Pangu-Weather apart is that it's outperformed the old-school, number-crunching prediction methods used by most weather services worldwide. This puppy was put to the test and came out a whopping "10,000 times faster" than the current top-dog forecasting tool, the one from the European Centre for Medium-Range Weather Forecasts.

The magic of Pangu-Weather lies in its focus on medium-range forecasting, looking forward up to two weeks. This type of forecasting is crucial for dodging disasters. Up until now, AI hasn't been able to outdo the traditional number-based prediction for this medium- and long-term forecasting, which basically works by breaking down weather conditions into little squares on a map and then playing the numbers game to predict changes.

Pangu-Weather’s new trick is its ability to handle the Earth's quirky 3D weather data by adapting to our planet's coordinate system. The Huawei team also used new computational strategies that do fewer iterations (repetitions of a process), cutting down on mistakes that add up over time. They trained the model using weather data from 1979 to 2021, repeating the process (an epoch) 100 times, each time using hourly weather data.

Impressively, Pangu-Weather can whip up a 24-hour global weather forecast in just 1.4 seconds using a single graphics card, which is a heck of a lot faster than the old number-crunching way. What's more, this tool isn't just fast, it's also pretty accurate. It's been put through its paces with tricky extreme weather events like typhoons and did a bang-up job. In fact, Pangu-Weather correctly predicted the path of Typhoon Mawar, this year's strongest tropical storm so far, five days before it showed up.

READ THE ARTICLE ON SCMP

Gamercraft raises $5 Million for AI-powered skill- gaming platform

Gamercraft, a gaming company from Miami, just got $5 million to level up their gaming platform. They're gonna use that cash to boost their tech, create new games, and amp up their marketing game. Big names like Alienware co-founders and others chipped in for this funding round.

Gamercraft is all about skill-based gaming where players can bet on their own matches. They're even cooking up ways to keep cheating to a minimum. They've already got a following of over 300,000 gamers from North America, Latin America, and Europe.

This isn't just your regular gaming platform though. Gamercraft is mixing things up with a combination of AI and blockchain tech, which allows for transparent and unchangeable records of transactions. This means they can keep track of player performance and dish out insights, using AI to make sure it's all fair play.

READ THE ARTICLE ON VENTUREBEAT

Wildfire detection startup Pano AI extends its $20M Series A with another $17M

Wildfire detection company, Pano AI, has added $17 million to its Series A funding, building on its original $20 million. Their smart system uses high-def cameras that constantly scan for fire signs. If the AI spots trouble, it rings the alarm bell. Pano AI is ahead of the game, getting the news out even before the first 911 call hits. The cameras are placed strategically so they can cross-check fire spots and give accurate coordinates, something a 911 call can't provide.

Pano AI's system allows emergency responders to get a jump-start on wildfires or keep expensive equipment idle if there's no real danger. It's like having a scout on the ground 24/7. They're all about giving the firefighters a head-start.

Despite a dip in the venture capital market, the company has pulled in this extra funding at a higher valuation than before, with the likes of Valor Equity Partners, T-Mobile Ventures, and Salesforce chipping in. The tie-in with T-Mobile is handy because Pano AI uses their 5G network and even their masts for camera placement.

Pano AI's customers are a mix of power utilities, private landowners, and government fire agencies. They don't sell the cameras, instead, they keep the hardware and rent out the software. The typical cost for this service? About $50,000 a pop, per year. The company's growth isn't slowing down, with plans to add hundreds more units by year-end, on top of the existing 100. Pano AI has its eye on both the U.S. and Australia and is also starting to look towards Europe.

READ THE ARTICLE ON TECHCRUNCH

AI transforms the humble chest X-ray into a better diagnostic tool

Researchers have jazzed up your typical chest X-ray with some fancy AI, making it a better tool for spotting heart issues. Normally, to really get a handle on heart health, doctors have to use an echocardiogram - or 'echo' - to check how well the heart's pumping and if its valves are in good nick. If not, that can spell serious trouble, like heart failure or even sudden death. But echos need a specially trained tech to run them.

That's where a team from Osaka Metropolitan University stepped in, training an AI to jazz up the common chest X-ray. This smart system was schooled using over 22,000 chest X-rays linked to their matching echos, taken from about 17,000 patients across four clinics. The researchers then tested the AI, finding it could spot six kinds of heart valve disease pretty darn well.

This newfangled AI system could be just the ticket when doctors need a fast diagnosis or don't have enough techs on hand. It might even be used when no specialists are around, during late-night emergencies, or for patients who can't handle an echo.

Lead researcher Daiju Ueda reckons this could be a game changer in terms of helping doctors diagnose heart issues and fill in the gaps where specialists aren't available. This groundbreaking work was published in the medical journal, The Lancet Digital Health.

READ THE ARTICLE ON NEW ATLAS

OpenAI’s Mysterious New Headquarters Features Nap Rooms And A Two-Story Library

OpenAI, the world-renowned AI company, has moved into a swanky four-story, 59,000 square foot headquarters in San Francisco's Mission District. They finished a ritzy $11 million renovation and expansion earlier this year, and they're keeping the details hush-hush.

We've caught wind of some cool stuff inside though, thanks to some peeped architectural plans and documents. This place has got it all: open space, snooze zones, mini-kitchens, lounges where you can catch both the sunrise and sunset, and a towering two-story library. There's even a reception desk made from cool burl wood, a "walk in the park" area flanked by live indoor trees, and a place for a giant 20-foot screen.

But OpenAI ain't just playing at home, they're also going global and plan on opening an office in London. What we know about this new HQ might just be the tip of the iceberg, though, because the company's been pretty silent about the final build. But if the plans are anything to go by, OpenAI is pulling out all the stops.

READ THE ARTICLE ON BUSINESS INSIDER

That Google memo about having ‘no moat’ in AI was real — and Google’s AI boss disagrees with it

Well, ain't this a pickle? A few months back, someone from Google seemingly threw a bit of a stink bomb, saying Google has no strong edge ("no moat") in the AI game. It seemed like a big deal, but now Google's AI head honcho, Demis Hassabis, has confirmed the memo's authenticity, but shrugged off its gloomy predictions.

The leaky memo, found in a public chat room, basically said that Google and its pal OpenAI ain't got the goods to top the AI leaderboard. In its place, a third player, open-source AI models, is sneaking in for the win. According to the memo, these upstarts are speedier, more flexible, more private, and basically doing a better job for the money.

But Hassabis, chief of Google's DeepMind, ain't buying it. He's pretty bullish about Google's future in AI. He thinks the company's competitive researchers and the merging of Google Brain and DeepMind (under his leadership) will bring more wins to the table. He points to the past and says, "Just look at what we've done before, and I bet we'll keep doing it. Heck, I reckon we'll do even better in the coming decade." So, for him, it's onwards and upwards, folks.

READ THE ARTICLE ON THE VERGE

Senators to receive first ever classified briefing on artificial intelligence

Senators are getting their first-ever peek behind the curtain on artificial intelligence. This hush-hush meeting, set for 3 p.m., is called by the big shots, like the National Intelligence Director Avril Haines and some other top-level folks. They're gathering in a super-secure part of the Capitol.

The big aim? To learn about how the U.S. is using AI to keep the country safe and to suss out what the competition's up to in the AI world. This briefing is just part of a series organized by a mixed bag of senators, including both Democrats and Republicans. So, it's about time our lawmakers get in the know about AI, don't you think?

READ THE ARTICLE ON THE HILL

Generative AI could add up to $4.4 trillion annually to global economy

Artificial Intelligence, or AI, can help make big decisions based on data and free up time, getting work done faster. Generative AI, a type of AI, could really shake things up, giving companies and their employees more time to focus on important stuff. It might even replace some of the work we do every day.

McKinsey, a big-name consulting firm, did a deep dive and found out some eye-opening things. They believe generative AI could pump between $2.6 and $4.4 trillion each year into the world economy. For context, that's more than the whole UK made in 2021!

Here's something that might make you sit up straight: generative AI could take over tasks that we spend 60-70% of our time on. That's because AI's getting better at understanding human language, a big part of many jobs.

In fact, if things keep going this way, half of all work could be automated by somewhere between 2030 and 2060, a decade earlier than expected. That could mean big productivity boosts but also means we need to help workers adapt.

But before you get too excited, remember we're just at the starting gate with this AI stuff. There are still risks to manage, and we need to figure out what new skills workers will need. And most importantly, we need to figure out how to help people transition to new roles or tasks.

READ THE ARTICLE ON ZDNET