Boost Your Game: Powerful AI Upgrades from Google, Amazon, Meta & HuggingFace

Today:

10 Badass AI Features Coming to Your Google Apps, Some of Which You Can Use Right Now

Looks like Google's cooking up a storm in the AI kitchen, folks. Ten new features are coming to a Google app near you, and some are already here, hotter than a jalapeño in July.

• First off, Bard, Google's gabby AI, is now free for all. Plus, it's getting a gloomy makeover with dark mode, can spit out text straight to Google Docs or Gmail, and now chats in Japanese and Korean. They're also throwing in image prompts and Google Lens integration. And more languages? 40 of 'em!

• They're also jazzing up their search with "Search Labs". It's not a lab coat, beaker affair, mind you. Just a way to try out new AI enhancements. Keep an eye out for a beaker icon to join the party.

• Working like a dog? Google's got you. Duet AI for Google Workspace is gonna be your new best friend. It helps you with everything from writing in Docs to designing in Slides. And forget form letters, now we have AI letters. The future's here, and it's oddly convenient.

• Ever wanted to compose music without learning an instrument? Meet MusicLM. Feed it a prompt and it'll spit out a tune. Be prepared for some surprises though - I asked it for 'German death reggae' and it drew a blank.

• Project Tailwind is an AI notebook that can summarize texts, create quizzes, and study guides. Sounds like a high schooler's dream.

• Google Maps is getting snazzy with Immersive View. It takes your regular map and turns it 3D. You can check traffic, air quality, and even weather conditions.

• Magic Editor is the new kid in town. It'll let you rearrange objects in your photos and fill in the gaps. Only on Pixel for now, but it'll probably hit the big time soon.

• Bored of your texts? Magic Compose can zhuzh them up, even in Shakespearean style if you fancy. Coming to the Messages app this summer.

• Android's not letting Apple have all the fun. Android 14's getting an emoji wallpaper generator and lock screen customization. Plus, Cinematic wallpaper brings your still photos to life

• Lastly, Find My Device now helps you locate not just your device, but also Bluetooth-connected goodies from other brands.

So, that's the lowdown. Google's going all-in on AI. Let's see if it's a full house or a busted flush.

READ THE ARTICLE ON GADGET HACKS.

Meta announces generative AI features for advertisers

Meta's rolling out a new playground, an AI Sandbox, to help advertisers spruce up their Facebook and Instagram ads. These marketers can now spin different versions of the same ad blurb for different crowds, all while keeping the gist of the message intact. Pretty nifty, huh?

They've also tossed in a feature to whip up diverse ad backgrounds faster than a New York minute. And if that's not enough, they've got a neat tool to help chop and change images for all kinds of uses, from your standard social post to snappy short videos.

For now, only a handful of lucky advertisers have the keys to the Sandbox. But come July, Meta plans to throw the doors wide open to more advertisers. So if you're feeling left out, hang tight.

This ain't out of the blue, though. Meta's CTO, Andrew Bosworth, hinted last month that they were itching to exploit generative AI tech for ads. Turns out, he wasn't just whistling Dixie.

This news follows Meta's Q1 2023 earnings report, which had them grinning like a Cheshire cat. They beat the big-wigs' predictions and saw a revenue growth for the first time in three quarters. While the tech giant's knee-deep in AI tools, they're still gunning for that metaverse development. So don't go thinking they've lost sight of the big picture.

As Meta's dipping its toes into the generative AI pool for advertisers, some ad tech startups are diving headfirst. Omneky, for instance, is using OpenAI’s DALLE-2 and GPT-3 to whip up ads, while Movio is using the same tech to create marketing videos. Guess everyone's keen on this AI bandwagon.

READ THE ARTICLE ON TECHCRUNCH.

Amazon reportedly has more AI features in development for Astro robot

Well, strap in folks, 'cause Amazon's Astro robot is getting an IQ boost. Turns out, the e-commerce big shot is stirring up some new AI tech to make Astro a better chatterbox. They're calling this upgrade "Burnham", and it's all hush-hush for now.

Remember, Astro's that home-monitoring bot Amazon threw at us a year and a half ago. Back then, it was all about being your pet's best friend and raising the alarm if you left the window open. Now, they want it to learn from watching us. Kinda creepy if you ask me, but hey, it's the future, right?

Currently, Astro is playing hard to get, only available by invitation and just in the U.S. And it doesn't come cheap. We're talking $1,599 for any of the three setups. There's the plain Astro, the Astro with Alexa Together for keeping tabs on your grandma, and Astro with Ring Protect Pro for catching suspicious folks in the act.

Amazon's big cheeses are pretty chuffed about the whole thing. They've got big plans for Astro and this Burnham tech. They're even talking about a whole lineup of smart robots.

They've been making strides in AI, launching their Bedrock service and teaming up with a company called Hugging Face. I know, sounds like a kids' show, right? It's actually pretty serious business. It's all about improving how businesses use AI models.

Amazon's top dog, Andy Jassy, is all in. He's excited about the potential of AI to change... well, pretty much everything. He let that slip during their recent earnings call. So, it looks like we're in for a lot more robots and AI in the coming years. Buckle up, it's gonna be a wild ride.

READ THE ARTICLE ON YAHOO! FINANCE.

HuggingFace Releases Transformers Agent

Alright, buckle up folks, I'm gonna break down the Transformers Agent for you. It's a new API that talks to a bunch of tools. Think of it as a multi-tool Swiss army knife for language processing. It's experimental and may change on a whim, so don't get too attached just yet.

This thing can handle a range of tasks, like generating images or reading text out loud. You tell it "Caption this image", or "Read this text out loud", and like a well-trained dog, it does it. It can even answer questions about a document if you ask it the right way. Pretty neat, huh?

To get this show on the road, you gotta install a few extras and set up an 'agent'. An agent is essentially a big brain that can handle different types of models. You can go with an OpenAI model, but that ain't free. Or, you can use BigCode or OpenAssistant which are free, but may not be as sharp.

Once your agent is set up, you're good to go. You can start running commands, like "Draw me a picture of rivers and lakes", or "Transform this picture to add an island". It can handle one or several tasks, but don't push your luck with complicated instructions.

The fun part is, you can also have a chit-chat with the agent. You can ask it to generate a picture, then add a rock to the picture, all in one conversation. You can run these commands on any device, no fancy hardware needed.

So, what's going on under the hood? Well, the 'agent' is a big language model, and the 'tools' are simple functions that help it understand what you're asking. The code it generates is safe to run, but if you're feeling paranoid, you can have the agent just return the code and decide for yourself whether to run it or not.

The agent has a bunch of built-in tools, like answering questions from a document, captioning images, converting text to speech, and translating text. You can also add custom tools if you're feeling adventurous.

To sum it up, the Transformers Agent is like having a handyman who can do a bit of everything, from painting pictures to reading text. You just gotta tell it what to do in a way it understands. And remember, it's still experimental, so don't expect it to fix your leaky faucet just yet.

READ THE ARTICLE ON HUGGINGFACE.

H2O AI launches H2OGPT and LLM Studio to help companies make their own chatbots

H2O AI, a California tech company, has just rolled out two new open-source tools. One's a fancy chatbot-making AI called H2OGPT, and the other's a no-code framework named LLM Studio. In plain English, they're giving businesses tools to build their own custom chatbots, no coding experience required. Pretty slick, right?

This comes as more businesses are showing interest in generative AI models – that's the tech that can generate human-like text. But there's a hitch. These companies are skittish about sharing sensitive data with big AI providers. And sometimes, these big providers can't really tailor their services to meet unique business needs.

So, what do H2OGPT and LLM Studio do? Well, they offer a sort of build-your-own-chatbot kit. Users can pick and choose different pieces of code, data, and AI models, then start building a chatbot that's just right for them. It's like building a teddy bear at Build-A-Bear, but for AI chatbots.

H2OGPT, H2O's own chatbot model, can also be used in commercial products. It's kind of like OpenAI's ChatGPT, but it's got a neat feature that lets users ask why it gave a certain answer.

These new tools are shaking up the AI world. Some folks are all for this open-source approach, while others, like OpenAI, are sticking to a more closed-off strategy. They're keeping their AI's training info under wraps, claiming it's for safety reasons.

H2O's CEO, Sri Ambati, acknowledges that AI can be misused, but he believes more people want to do good with it. He suggests using AI-driven checks to prevent misuse.

In the end, it seems like H2O AI is all about giving power to the people. Or in this case, to the companies wanting to get their hands dirty with AI, without needing to be coding whizzes. But remember, with great power comes great... chatbot-building capabilities? I think that's how the saying goes.

READ THE ARTICLE ON VENTUREBEAT.

Google will label fake images created with its A.I.

Alright, folks. It seems Google has been up to some shenanigans. It's going to start tagging those computer-made images floating around the internet, you know, the ones that can sometimes fool your Uncle Jerry into thinking the Pope's gone hipster.

Yep, Google's got a new plan. They're going to hide some secret info inside these AI-generated images. No, you won't see it, but Google's other software like its search engine will. It'll then slap a big ol' warning label on those images, like a scarlet letter for artificial images.

And it's not just the warnings; Google's throwing in some extra info to keep folks honest. Things like when the image first popped up online and whether any news sites have been chatting about it.

They're calling it "generative AI", pretty fancy, huh? But the whole purpose is to stop these images from helping the bad guys, you know, spammers, scammers, and folks pushing fake news.

Now, you might ask, "Can't they just spot these AI images?" Well, it's not that simple. Sure, there might be some hints like botched hand drawings, but there's no foolproof way to tell if an image was birthed by a computer or a human.

So Google's game plan? Tag 'em as they're created, not afterwards. Seems sensible. Even got some support from Shutterstock and Midjourney.

Oh, and by the way, Google's got some new toys too. They just announced a $1,799 folding phone and some extra AI stuff for their other products at their annual powwow. Sounds like they've been busy, huh?

READ THE ARTICLE ON CNBC.

AI2 is developing a large language model optimized for science

Alright, let's have a yarn about AI2, a nonprofit institute that’s decided to dive into the deep end of artificial intelligence (AI) language models. They're cooking up a new AI model named OLMo, or the Open Language Model, that aims to be more open than a 24/7 diner, granting researchers full access to see what's under the hood. It's like inviting someone to check out your secret BBQ recipe – a bit risky, but it could make the world a better (or at least, more delicious) place.

OLMo is a group project, with tech heavyweights like AMD and a supercomputer-loving consortium called LUMI pitching in. MosaicML and Surge AI are also on board, supplying training code and data, because apparently, you can't make an AI language model with just some duct tape and good intentions.

Their goal? Simple: create the best open language model in the world. Lofty ambitions, but at least they're shooting for the stars. They're keen to foster a real "all for one, one for all" kind of vibe, making all aspects of OLMo widely available for others to use or tweak.

Interestingly, OLMo is supposed to be a big bookworm, focusing more on textbooks and academic papers instead of code. A sort of nerdier cousin to other AIs, if you will. It's like they're trying to create a model that could win at Jeopardy, but only if all the categories were scientific research papers.

As for the knotty issues around legality and ethics of AI? Well, they say they're going to work with AI2's legal department and some yet-to-be-named experts, stopping at various checkpoints along the way to hash things out. Sounds a bit like a road trip, but with more legal jargon and fewer gas station snacks.

There's also some mumbo-jumbo about an ethics review committee and a plan to limit access to certain parts of the model to keep the bad guys at bay. Guess they're trying to keep the AI equivalent of a Pandora's Box from causing havoc.

Anyway, they're set to start training this brainy beast on a supercomputer in Finland (apparently Europe's fastest) in the coming months. In the meantime, they're looking for more folks to join the party and chip in on the project. So, if you've got a knack for AI and a hankering for open-source projects, maybe you should give them a ring.

READ THE ARTICLE ON TECHCRUNCH.