Microsoft's Phi 1.5 Unlocks Big Power in Small Models

Today:

Microsoft pushes the boundaries of small AI models with big breakthrough

Microsoft's research team has jazzed up one of its smaller AI models, making it a cool, budget-friendly option with some snazzy features like OpenAI’s big kahuna, GPT-4. Their model, Phi 1.5, is now a jack of all trades, handling images and text without bulking up in size. This could be a game changer, making AI tech more accessible and taking the heat off the heavy demand for fancy graphics processors.

Unlike GPT-4, which is a real power hog and has a whopping 1.7 trillion parameters, Phi 1.5 is more like a lightweight, rocking only 1.3 billion parameters. Imagine GPT-4 as the Empire State Building and Phi 1.5 as just a footlong sub sandwich—still packs a punch but way easier to handle.

Sebastien Bubeck from Microsoft Research is stoked, saying they've proven that you don’t need a gigantic model to do cool stuff with AI. This could be a real money saver, especially for businesses that use AI to chew through big chunks of data. While an average Joe might use ChatGPT to whip up an email, companies might be shelling out around $100 for every 1,000 prompts to dig into corporate data. With Phi 1.5, Microsoft is showing you can get the bang without the big bucks.

READ THE ARTICLE ON SEMAFOR

Matic’s robot vacuum maps spaces without sending data to the cloud

Matic's a fresh kid on the block, kickstarted by a brainy former Google dude named Navneet Dalal and his buddy Mehul Nariyawala. They're all about making robots that can cruise around your house cleaning floors like a pro, without phoning home to the cloud with your pad's layout.

These guys have managed to round up nearly $30 million from some big-shot tech names, like the folks who started GitHub, Stripe, Quora, and Twitter. Talk about having friends in high places.

Matic's vacuum, which they're just teasing now and ain't even on sale yet, is some kind of high-tech gizmo with cameras that can tell if it's mopping or vacuuming and even remember where it's been to get smarter every time. They're letting friends and some lucky customers give it a whirl before it hits the market.

READ THE ARTICLE ON TECHCRUNCH

Kaiber’s new app helps artists create music videos using generative AI tools

Kaiber's shaking up the music video scene with a fresh app that lets any Joe or Jane make sick videos with AI—no big bucks needed. This brainchild of Kid Cudi and Linkin Park's video creators lets you whip up animations by typing out your dream scene or using your own pics and clips. It's got styles for days, from old-school flipbook looks to smooth moves that flow like butter.

You can mess with the camera angles, fit your vid to any social media shape, and drop your tunes in there. The app listens to your music and makes the animations groove to it. Plus, if you're low on cash, you can use the app for nada, but your videos will carry the Kaiber tag.

They've also got this sweet deal where you make videos using beats and vibes from some indie artists like Oksami. Got a following? Kaiber will pay you to bring them new peeps.

It's on your phone, no matter if you're an Apple fan or an Android head, with plans that won't break the bank. You start with a free taste and then decide if you want to dive in deeper.

READ THE ARTICLE ON TECHCRUNCH

Turn Speech Into Text With Audio Hijack 4.3’s New Transcribe Block

Audio Hijack, the go-to app for recording anything audio on your Mac, just dropped a cool new feature that's basically like having a personal note-taker. It's called the Transcribe block, and it's part of their latest update, version 4.3. What it does is simple: it turns what you say into written words. You talk, it writes, and it understands a whopping 57 languages!

Now, the best part? It's totally free. No more paying through the nose for someone to write down what's said. And it's all private, happening on your own computer without throwing your words up to the cloud.

Whether you're chatting on Zoom, Skyping, or recording a podcast, you can transcribe all of it. Super handy for meetings or interviews when you don't want to be that person clacking away on a keyboard. And it's not just live stuff—it can transcribe recorded files too.

READ THE ARTICLE ON ROGUE AMOEBA

Factory wants to use AI to automate the software dev lifecycle

Two dudes, Matan Grinberg and Eno Reyes, cooked up an AI platform at a San Francisco Hackathon. This thing can automatically tackle basic coding tasks. They thought it was pretty nifty, so they expanded it, founded a company called Factory, and decided to make some green off it. Their main pitch? Making software engineering way easier by letting AI handle the grind of tasks like code review, updating docs, and testing.

Factory's AI systems, nicknamed “Droids” (yeah, like Star Wars), can do stuff like giving code reviews, fixing up code, and even drafting new code. Grinberg mentions they’re built on a “Droid core”, which basically learns from a company’s tech data and uses it to solve problems.

Grinberg admits they don’t have the dough to create all their AI models themselves, so they’re kinda riding on third-party providers. But the dream? Factory wants to eventually build its own top-notch AI system using data from its early bird customers.

READ THE ARTICLE ON TECHCRUNCH

Five takeaways from UK’s AI safety summit at Bletchley Park

Here's the lowdown on the UK's AI safety shindig at Bletchley Park:

UK's Big Win:

The UK Prime Minister, Rishi Sunak, scored big with this summit. It was like throwing a party and having all the cool kids show up: the US VP, the head honcho of the European Commission, and even Elon Musk. They signed this pledge thing saying, "Yeah, AI could be sketchy, we should watch out," which is a pretty solid move.

USA Flexing Muscles:

The US showed they're boss when it comes to AI. Biden's like, "You wanna play with AI? Show me what you got first." Plus, VP Harris is all about starting an AI safety club. The UK's chill with it though; they know the US has the upper hand with all the big AI firms.

Elon Musk Steals the Spotlight:

Elon Musk showed up and, man, did he make waves. He's got this vibe that turns heads, even though his AI game isn't as big as others. The Prime Minister chatted with him on a live stream and Musk's doomsday AI talk got more buzz than probably deserved.

Everyone's Freaking Out Differently:

There's this split at the party: some folks are sweating bullets over AI ending the world, while others are more worried about next year's elections getting messed up by fake news bots. There's no handshake agreement on the end-of-the-world stuff, but everyone's pretty much on the same page about keeping elections legit.

Different Strokes for Different Folks:

Every country's trying to one-up each other on who's the boss of AI rules. Europe's all, "Been there, done that," and the US is talking big about their new AI safety squad. The UK's playing it cool, saying "Slow down, it's too early for rules." But despite the chest-thumping, everyone's into these global get-togethers because, hey, it's better to chat about the AI drama together than alone.

Bottom line: The summit's got everyone talking, agreeing we gotta keep our eyes peeled on AI, and maybe, just maybe, start agreeing on how to handle it.

READ THE ARTICLE ON THE GUARDIAN

The most in-demand AI job of 2023 can pay over $200,000 and offers remote opportunities

People are going nuts for AI jobs since ChatGPT came around last year. Job searches for AI gigs on Indeed have shot up like a rocket — think 4,000% more in just a year. And even though there's a bit of a dip from last year's frenzy, there are still loads of AI jobs out there, more than before the world got hit with the pandemic.

The big job everyone wants isn't building the AI — it's understanding the data it spits out. That's right, being a data scientist is where it's at. They're the rockstars in the AI job market, making sense of all the numbers and helping companies get smarter in how they do business.

And it's not just tech companies hunting for these number crunchers. Banks, rocket shops, media — all sorts of places are jumping into the AI pool and need data scientists to help them swim. A year or two ago, companies were just flirting with AI, handing out temp jobs like candy on Halloween. But now, they're going steady, offering full-time spots because they see how AI can really amp up their game.

READ THE ARTICLE ON CNBC

How AI detectors can destroy innocent writers' livelihoods

Michael Berben, not his real name, got the short end of the stick. This guy's been writing his heart out for years, dropping 200 articles as a freelancer. Then, out of the blue, his main gig drops a bomb — says his stuff rings like it’s penned by a robot. They're using this AI detector that's hollering "95% chance of AI work" on his latest pieces. Even the old stuff, pre-robot-era writings, is getting flagged. No more gigs, just like that.

OpenAI, the ChatGPT folks, ditched their own 'cause it couldn’t tell its left from its right. Writers like Mike could write in a way that trips these detectors, even though they're doing the legwork old school.

AI detector companies are blowing their trumpets, selling folks a dream with "99% accuracy" badges. But it’s all smoke and mirrors. It's like those skincare ads with the tiny print saying "based on 40 people." That ain't the real world.

End of the day, Michael's story doesn’t have the Hollywood ending we all root for. His client was so spooked by the chance of Google giving AI-written content the cold shoulder that they wouldn't budge. So, he’s out of a job, courtesy of some AI detector playing umpire and calling the wrong shots.

READ THE ARTICLE ON AUTHORY