OpenAI Assistants API to Build AI Agent SWARMS

Today:

OpenAI Assistants API to Build AI Agent SWARMS. Better than AutoGen?

Recently, the OpenAI Developer Conference went down, and now they're sharing videos from it with everyone. They're showing off some neat stuff, like how to make AI assistants using API calls. The big news from the conference was about these two things: GPTs and assistants. They're both about tweaking the latest OpenAI model, GPT-4 with vision, to make your own chatbots or apps.

If you're itching to create your own AI helpers, the way to go is using API calls. This is less restricted, letting you do more stuff. You don't have to be a pro coder to jump into this AI game.  It's all about learning as you go, making your own AI app in no time.

WATCH THE VIDEO ON YOUTUBE

DeepMind and YouTube release Lyria, a gen-AI model for music, and Dream Track to build AI tunes

Google's DeepMind and YouTube just dropped a game-changer in the music world called Lyria. It's a new AI music-making tool that works hand in hand with YouTube. They've also introduced Dream Track and Music AI, two fancy tools that help creators make music from simple stuff like humming a tune. They're even adding a special watermark for AI music, just like they do for AI images.

Dream Track is rolling out to a few creators first. They can pick an artist, and the AI whips up a 30-second track in that artist's style. These tracks are meant for YouTube Shorts. And guess what? Real artists like Charlie Puth and Demi Lovato are in on it, helping to test and improve these models.

YouTube's music bigwigs, Lyor Cohen and Toni Reed, mentioned that their Music AI tools are coming from a group of artists and producers who are all about pushing the creative envelope with AI.

While Dream Track is just starting out, the full suite of Music AI tools will be available later this year. These tools are wild! You can hum a tune and it'll create music with specific instruments, or even a whole band. You can start with a simple hum or a vocal line, and the AI builds the rest of the track around it.

READ THE ARTICLE ON TECHCRUNCH

Meta teases new AI-powered editing tools coming to Facebook and Instagram

Facebook and Instagram are about to get super cool with new AI toys for editing pics and making videos. Mark Zuckerberg, the big boss at Meta, just spilled the beans on two new tricks they've got up their sleeve, all thanks to this fancy AI tech called Emu.

First up is "Emu Edit." This thing is a game changer for tweaking photos. Say you've got a pic but want to ditch or swap out something in it, like a photobombing squirrel. With Emu Edit, just type what you want changed, like "make that squirrel a hot air balloon," and bam, it does it without you needing to be a Photoshop wizard.

Then there's "Emu Video." This one's about making videos from just a few words or a picture. You tell it what you want, and it whips up a video. Sure, it's not Hollywood-level, but it's way better than the choppy stuff we saw from Meta last year.

We don't know when these cool tools will hit Facebook or Instagram. Meta's keeping mum on the release date and whether this is the same stuff they hinted at last year. 

READ THE ARTICLE ON THE VERGE

Instagram adds new features, including custom AI stickers, photo filters, a clip hub and more

Instagram is rolling out this cool feature where you can make your own stickers. You just pick a photo or video from your phone or Instagram and the app's AI cuts out whatever you want to turn into a sticker. It's kinda like what Apple did with iMessage in iOS 16, but Instagram's version lets you adjust things if you're not happy with the AI's first try.

There's also this other AI sticker thing they showed off before, where you type in some wacky idea and it creates a sticker for you. That's been around for English users for a bit now.

Instagram's throwing in some new tweaks for creators on Reels – stuff like undo and redo, messing with your clips' size and angle, and a new hub for making memes with audio. They're also bringing back photo filters, adding new text-to-speech voices, fresh fonts, and some outline options.

For those who like to keep track of their stats, Instagram's mixing up how they count views on Reels. They're adding this "Replays" metric to the usual "Initial Plays," so now you'll see a total "Plays" count. It might pump up your numbers, but it's still handy to see how many different folks are watching.

These updates are hitting Instagram starting now, but they might take a little while to show up for everyone. So keep an eye out!

READ THE ARTICLE ON TECHCRUNCH

Spotify’s podcast and audiobook discovery will get a boost from Google Cloud’s AI

 Spotify's hooking up with Google Cloud to sharpen up how folks find podcasts and audiobooks on their platform. Think of it like a super-smart system that digs deep into about 5 million podcasts and a whole bunch of audiobooks, fine-tuning the info about them so you find exactly what you're into.

Also, they're on the lookout for sketchy content. Spotify's got rules against certain types of no-go audio, and they're tapping into this AI magic to keep things clean, mixing robot smarts with real people making calls.

Spotify's already playing with AI for tunes – they've got this thing called AI DJ. Mixed reviews, but it shows they're serious about getting techy with not just music, but podcasts and audiobooks too. They're all about making sure when you hit play, it's something you're actually gonna dig.

READ THE ARTICLE ON THE VERGE

Microsoft launches a deepfakes creator at Ignite 2023 event

Microsoft rolled out a fancy new tool that lets you make a photorealistic avatar that can chat away, saying stuff the real person never did. It's called Azure AI Speech text-to-speech avatar. You feed it some photos, type out a script, and boom - you've got a talking avatar. It can even mimic someone's voice with the right training.

Microsoft is pitching this for making training videos, product intros, and more, just by typing what you want the avatar to say. These avatars can speak different languages and even chat back using AI models like OpenAI’s GPT-3.5.

They've got another tool coming out called personal voice. It can clone your voice from just a minute of audio. Great for personalized assistants or dubbing content, but they're being cautious. They're demanding explicit consent, and you can't use it for just any content.

Microsoft's adding watermarks to these AI voices to tell 'em apart from real ones. But to detect these watermarks, you need Microsoft's approval to use their detection service. So, it's a bit of a mixed bag.

READ THE ARTICLE ON TECHCRUNCH

You can now use AI in Google to help with your holiday shopping. Here’s how

 Google's rolling out some cool AI features for holiday shopping. Say you're hunting for a gift for your 4-year-old nephew who's into science, Google's got your back with some solid suggestions. They're really stepping up their game with this AI magic, especially with the holidays coming up and folks ready to spend big.

Here's the breakdown of what's new:

• Gift Finder: Just type what the person's into, like "gifts for home cooks," and Google serves up a list of stuff they might dig, even things like cooking classes.

• Picture This: Describe what you're looking for, maybe a "puffy, colorful metallic jacket," and Google whips up a realistic image, then shows you real products that match.

• Virtual Try-On: Search for something like a "brown jacket at J-Crew," and Google shows it on 40 different models of various sizes and backgrounds. It's like seeing how it'd look on different folks.

You gotta opt into these features, but they're pretty neat for finding the right gift or just exploring options. And hey, Google's not new to this game, but they're definitely upping the ante this holiday season.

READ THE ARTICLE ON CNBC

Can AI solve medical mysteries? It’s worth finding out.

It's tough as nails finding a doc these days, so folks are hitting up Google for their health woes. Even someone raised by scientists is googling symptoms. There's a new player in town, though: chatbots using big brain AI to dig through medical info and spit out easy-to-get answers. They're even nailing some tricky diagnoses, like this kid who stumped 17 docs but got sorted by a chatbot.

People might start trusting them more than real docs, or the AI could start cooking up bogus health tips. Especially for rare diseases, affecting millions, AI's got game. Docs are ace with common stuff, but there are thousands of rare diseases they might not even know.

There's this Undiagnosed Diseases Network, where they're training AI to sift through patient records to speed up diagnoses. Plus, AI could give patients a voice when docs aren't listening, like this mom-doc whose concerns were brushed off until her kid was diagnosed with a rare disease years later. AI's also helping link up patients with the same rare mutations, sparking hope for research and treatments. 

READ THE ARTICLE ON THE WASHINGTON POST

AI boosts breast cancer detection rates, new research shows

AI's making some big moves in catching early-stage breast cancers that docs might miss. We're talking about spotting up to 13% more cases than the usual screenings—pretty major since about 20% of cancers can slip through the cracks with the old methods.

They used this AI tool called Mia, cooked up by Imperial College London and Kheiron Medical Technologies. They tested it on 25,000 women in Hungary from 2021 to 2023. The study had three parts, each tweaking how radiologists and AI worked together. The results? Cancer detection rates jumped by 5%, 10%, and even 13% compared to the usual double-check by radiologists.

Most of the extra cancers found were invasive, the kind that can spread. So, catching them early is a big win. Another study from Sweden found similar results, showing AI can really keep up with human double-checking.

Mia's already being tested in 16 UK hospitals and is starting to roll out in the US. The researchers are pushing to expand this kind of study to more places and different AI systems. They also want to keep an eye on how these detected cancer cases evolve.

READ THE ARTICLE ON FINANCIAL TIMES

Several popular AI products flagged as unsafe for kids by Common Sense Media

Common Sense Media, the crew known for rating movies and games for families, just dropped a new kind of rating: AI products. They're doing this because parents are kinda in the dark about whether AI stuff like ChatGPT or Snapchat’s My AI is cool for their kids. Most parents (82%) are itching for some guidance, but only 40% know where to find it.

So, Common Sense Media checked out 10 big-name AI apps, including learning tools and chatbots, and rated them on stuff like trust, safety, privacy, and how much they can teach kids. Turns out, the generative AI ones, like DALL-E and Snapchat’s My AI, didn’t do so hot. They got dinged for potentially teaching kids biases or giving off weird vibes, like reinforcing stereotypes or privacy issues.

The AIs that got gold stars? Educational ones like Ello’s AI reading tutor and Khanmingo. They’re not as flashy as the others, but they’re designed with kids in mind and focus on being fair and clear about privacy.

Common Sense Media plans to keep these reviews coming, hoping to enlighten not just families, but lawmakers and all the big wigs, about what’s cool and what’s not in the AI world. They're pushing for AI to be safe and transparent, especially for the youngsters.

READ THE ARTICLE ON TECHCRUNCH