- NATURAL 20
- Posts
- ChatGPT Meets iPhones, Meta Crafts AI Chip: A Giant Leap in AI
ChatGPT Meets iPhones, Meta Crafts AI Chip: A Giant Leap in AI
With ChatGPT's grand entry into the iOS world, iPhones are now more conversational than ever. Simultaneously, Meta's unveiling of their ground-breaking MTIA chip promises a new dawn for Artificial Intelligence.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/151875c2-789e-4e16-b5e7-1737bdf68f75/natural_20_banner.png)
Today:
OpenAI introduces ChatGPT app for iOS, bringing most popular AI chatbot to iPhones
Looks like OpenAI's been cookin' up something good for all you Apple fans. They've gone and dropped a ChatGPT app for iOS, meaning iPhones are now getting a serious dose of AI mojo, just a half-year since ChatGPT first wowed us all.
Introducing the ChatGPT app for iOS! We’re live in the US and will expand to additional countries in the coming weeks. Android is next!
— OpenAI (@OpenAI)
4:51 PM • May 18, 2023
ChatGPT in the App Store is the real deal, according to OpenAI's blog post. It syncs up your chats, takes voice commands, and updates you with the latest AI bells and whistles. It's free, y'all! Even keeps your chat history tidy across devices, and lets you talk to it courtesy of Whisper, OpenAI's speech-recognition system.
For the fancy folks who subscribe to ChatGPT Plus, they get to tango with GPT-4's shiny new features and enjoy the fast lane with speedier response times.
Our man Sam Altman, CEO of OpenAI, might've been hinting at this when he tweeted, "summer is coming." Makes sense now. As knock-off ChatGPT apps have been sprouting up like weeds, and smaller devices showing they can handle the big league language models, OpenAI had to act pronto.
summer is coming
— Sam Altman (@sama)
3:19 PM • May 13, 2023
Details on safety? Eh, not so much. Just a vague promise to keep improving features and safety as they get feedback. But hey, they're on it.
Non-Apple users, don't you fret. OpenAI's blog post says ChatGPT will be making its way to Android soon. Buckle up, folks!
So, ready for a chit-chat with your AI buddy? Get the ChatGPT app here.
Meta is working on a new chip for AI
Well, butter my biscuit, looks like Meta's cookin' up something big - their first in-house chip built for running AI models, the MTIA. Apparently, it's gonna offer more juice than a CPU and will be tailor-made for Meta's own needs. Combined with GPUs, Meta's promising the whole nine yards - improved performance, decreased latency, and more efficiency. But don't start popping the champagne yet, this fancy chip ain't hitting the market 'til 2025.
Not content with one star player, Meta's also rustling up a special video-transcoding chip called the MSVP. Designed to support high-quality video and speedy live-streaming, it might eventually help bring AI-made, AR, and VR content to Meta's apps. Sounds like some futuristic jukebox, don't it?
There's more. Meta's crafting a so-called 'next-generation data center design', which they claim will be AI-optimized, faster, and cheaper to build. They're also tooting their own horn about the Research SuperCluster (RSC) AI Supercomputer, claimed to be one of the fastest in the world. Bold claim, considering every other tech giant out there seems to be in the same race. But as the saying goes, if you've got it, flaunt it. Guess we'll have to see if Meta's bite matches its bark.
AI-powered coding, free of charge with Colab
So, here's the long and short of it, folks: Google Colab, your favorite no-fuss, no-muss Python playground, is getting a facelift. Been around since 2017, it's served over 7 million folks - students, nerds, and everyone in between, all on the house. Good for everything from machine learning to data crunching to just learning the ropes.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/7bfcc4c5-70a2-4fff-8ac6-259a5a2aacb4/generate-cropped.gif)
Now, it's ready for a new trick. They're throwing in some fancy AI coding features – things like finishing your sentences but in code, translating plain English into Python (no Rosetta Stone required), and even a chatbot that might know more coding than you do.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/3f059dcb-8332-479f-8d59-cf334860d77c/chat-cropped.gif)
This new hotness is powered by Codey, the new cool kid on the block. Born from the latest and greatest PaLM 2 model, Codey's got a brain trained on heaps of high-quality code. They’ve also given this whiz kid a little extra tutoring for Python and Colab-specific tasks. So, buckle up and get ready to code faster than a greased lightning.
An open letter from co-founder and CEO of Zapier
So, Zapier's founders reckon that computers ain't working hard enough for us. With AI, they think that might change. It's a little like giving your old pickup truck a jet engine - now that'd get you to the store in a jiffy. They're thinking everyone, not just the techy whizz kids, should be able to create their own software apps.
Their aim? To hand over the AI tools to everyone, kinda like giving a kid a brand new box of LEGOs. No more 'one size fits all' business solutions. Now, you can tailor your apps like a suit, fitting snugly around your specific needs. Let's leave engineers to do the heavy-lifting while everyone else gets to have some fun, being creative with their own custom software.
Zapier customers are already riding the AI wave, turning their workflow into a smoothly automated dream. They're integrating AI tools like OpenAI, Leap, and Vowel into their Zaps, finding uses from cutting down paperwork to boosting their lead generation.
They've also released a ChatGPT plugin. Picture this - you can connect to apps and automate tasks right inside your chat. No more toggling between windows. Plus, their Natural Language Actions API can even give your AI model an automation layer.
@mike.rama Integrate AI into your business with @zapier This video walks through the steps of integrating Zapier with OpenAI and Twitter so you can ... See more
And here's the cherry on top - you can create a Zap with plain ol' English. No need to sweat over code. Just tell 'em what you want and they'll draft a Zap for you to tweak to your heart's content.
Got @HappyHarryAI running in iMessage and hooked up to @zapier NLA 👏
— Harrison Jackson (@HarrisonJackson)
10:57 PM • Mar 22, 2023
In essence, they're working to make sure you're not just a bystander in the AI transformation. Jump in, the water's warm. Go check out zapier.com/ai if you fancy being at the forefront of all this exciting stuff. It's like being the first one at a yard sale - who knows what treasures you'll find.
New Google AI tools will let advertisers generate media assets
Alright, folks, gather 'round the ol' digital campfire. Our pals at Google have been tinkering in their tech lab again. This time, they're whippin' up some AI tools that could shake up the ad game for those YouTubers and ad wizards out there.
These fancy gizmos, fueled by something called PaLM 2 (don't worry, no degree in rocket science needed here), might just be the ticket for creating media assets. Sure, the details are a bit thin on the ground, but that's the big picture.
In the past, Google's AI seemed more like a DJ, remixing what already existed. Now, it looks like it might start making its own beats, generating assets and whatnot.
It ain't just for Google Ads, mind you. Word on the street is that they've been test driving this PaLM 2 thingamajig on YouTube too. Maybe helping young creators brainstorm video ideas, generate titles, and all that jazz.
Why should you care? Well, Amazon and Meta are also dipping their toes into this AI pool. If Google's new toy can boost performance - maybe get those conversion rates up and costs per lead down - it's worth paying attention.
Eyes peeled for this year's Google Marketing Live on May 23, it's likely to be a goldmine for AI news. Will we get the full scoop on these AI tools then? We'll keep our ear to the ground and keep y'all posted.
Scientists Train New AI Exclusively On The Dark Web
All right, so here's the skinny: Some South Korean brainiacs went deep down the rabbit hole and taught an AI, affectionately nicknamed DarkBERT, to navigate the internet's shady back alleys – you know, the dark web. They didn't just wake up and choose chaos. Nope, they were looking to arm this new AI detective with the tools to fight the bad guys of the cyber world.
Transforming this AI into an eagle-eyed crime fighter wasn't a cakewalk. The folks had to connect it to the Tor network (that's basically the yellow brick road leading to the dark web), and let it snoop around to create a database of all the dirt it could dig up.
Now, these researchers are patting themselves on the back, saying their new AI Sherlock Holmes is better at decoding the gibberish of the dark web than previous models, including one Facebook had up its sleeve. DarkBERT, they reckon, could become a top gun in cybersecurity, sniffing out ransomware sales or data leaks, and keeping tabs on the dark web's seedy forums.
But let's hold our horses here. While it sounds promising, we're still in the "I'll believe it when I see it" phase. Plus, do we really want to hand the AI the sheriff's badge to police the web? Sounds a bit like letting the fox guard the henhouse, doesn't it?
A Meta Platforms Leak Put Powerful AI in the Hands of Everyone
Meta, the big dog running Facebook and friends, has decided to play Santa Claus with its shiny new AI tech, LLaMA. It's open-source now, so anyone who can cough up an email address gets the code after a quick check. Free AI for all!
"Open's the way to win," says Meta's AI bigwig, Yann LeCun. But not everyone's happy. Google and OpenAI reckon this open-source parade could lead to a bit of a mess. Think bots spewing out bad stuff or, worse, job-stealing automatons.
The hot-off-the-press LLaMA tech was on the infamous 4chan in a jiffy, prompting more than a few raised eyebrows. Google's AI gatekeeper, Zoubin Ghahramani, is like, "Where's this free-for-all headed?"
Meta ain't flinching though. LeCun retorts, "You want a couple of US giants to control all AI?" He's got a point.
Now, this ain't a new spat. Open source vs. closed source is an old tech tug of war. Most notably, Google used the open-source strategy to challenge Apple’s smartphone stronghold with Android.
And then there's LLaMA – Meta's brainchild and one of those fancy "Large Language Model" thingamajigs that can pen poems, code software, or chat your ear off. Open-sourcing it, including the training data, means folks can get to the AI party without shelling out big bucks on equipment. But, as some Stanford eggheads found out, you can end up with some nasty surprises. Even something as ghastly as an AI spewing Hitler-approved ideas.
LeCun’s got a counter though: "Sure, folks might create dangerous stuff, but we can keep it from going viral." Let’s see how that holds up.
But why this generosity from Meta? Well, it's a bit like judo. Get everyone using your tools and you become the heavyweight that can't be knocked off. And it has worked before, says LeCun, pointing at the success of the internet.
His take: "The more open, the faster we progress." In other words, let the AI games begin!
Pandas AI: The Generative AI Python Library
Alright folks, if you're into data science and love Python Pandas like a squirrel loves nuts, get ready for a big ol' shake-up. Now there's Pandas AI, a new library that lets you chit-chat with your data. Yeah, you heard right, no more hours lost to playing 'Where's Waldo?' with rows and columns.
What this means is, you can pop a question to your data and get a quick comeback. It's like a sidekick for your existing Pandas, not a replacement. It's about cutting down on time spent prepping data and getting straight to the juicy analysis part.
Now, don't get too excited, you'll still need to know your Python to make the most of it. With OpenAI API's help, Pandas AI is aiming to let you virtually chinwag with your machine to get the results you want. No more writing a bunch of code, just tell it what you need and it'll spit out the results in machine code.
Getting it set up is as easy as pie. Just a couple of installs, import a few things, and you're off to the races. You can even start asking it some fancy questions about your data and it'll spit back a DataFrame with the answers.
The team behind Pandas AI is just getting started. They're cooking up some more improvements like adding support for more LLMs, making PandasAI a CLI, creating a web interface, and adding unit tests. They're open to suggestions if you're feeling chatty.
So, if you're a data pro and want to give Pandas AI a whirl, don't be shy. Let us know how it works for you. Just remember, it's a tool to help speed things up, not a magic wand. You'll still need to know your stuff to correct it when it goofs up. Happy data conversing, y'all!
What'd you think of today's edition? |
What are MOST interested in learning about AI?What stories or resources will be most interesting for you to hear about? |
Reply