How to BUILD OpenAI Playground Assistant

Today:

OpenAI Playground Assistant is POWERFUL! Here's how to BUILD one…

OpenAI just dropped some seriously cool stuff for all the ChatGPT plus members out there. Now, you can create your very own AI assistant, and it's a game changer. This video is all about showing you how to whip up your assistant in just five minutes. You start by heading over to OpenAI's playground and pick 'assistant' from the options.

This tool is a massive timesaver. What used to take 20 minutes of tedious work can now be done in a couple of minutes for just a few cents. It's not just about saving time; it's about making things less of a headache.

WATCH THE VIDEO ON YOUTUBE

OpenAI Data Partnerships

OpenAI's got this new thing called Data Partnerships. They're teaming up with different groups to get all sorts of data to teach AI better. The goal? To make AI that gets the whole picture - people, what drives us, how we talk, you name it. They want AI to get a grip on everything, from different jobs to cultures and languages. By adding your stuff to the mix, the AI can get smarter in your specific area.

For instance, they worked with Iceland's government and a company called Miðeind ehf to beef up GPT-4's Icelandic. And there's this non-profit, Free Law Project, that's helping by adding a ton of legal documents. OpenAI thinks there are more folks out there who can pitch in with their unique data.

These Data Partnerships are a way for more groups to shape the future of AI and get AI that's more tuned to what they care about.

READ THE ARTICLE ON OPENAI

Nvidia develops AI chips for China in latest bid to avoid US restrictions

Nvidia, the big shot in AI tech, just cooked up three new chips specifically for China. They're doing this fancy footwork to dodge US export rules that are cramping their style. This isn't Nvidia's first rodeo, though. They had to rejig their gear for China before, thanks to Uncle Sam's new rules.

These new chips, called H20, L20, and L2, aren't as beefy as the old ones, but they still pack a punch for the Chinese market. Nvidia's playing it smart – they're keeping these chips cool enough to slide under the US radar but still hot enough to sizzle in China.

Nvidia's been a big fish in the AI pond, with their A100 and H100 chips being the go-to for building brainy bots like ChatGPT. But last year, the US put the brakes on shipping these top-tier chips to China. Nvidia's answer? They whipped up the A800 and H800, a bit less zippy but still solid. Then, bam! The US tightened the screws again, and even these had to be shelved.

With the new rules hitting fast, China's tech giants got stuck with older chips for their AI dreams. Meanwhile, Nvidia's not sleeping – they're already pushing these new chips out for testing, signaling they're about to hit the big time.

READ THE ARTICLE ON FINANCIAL TIMES

ChatGPT is powering a new kind of Snapchat lens

Snapchat's cooking up some new AR lenses that are gonna be powered by ChatGPT. They let the cat out of the bag about this at their Lens Fest, a big shindig for folks who make Snapchat filters. These aren't your ordinary filters, though. Think of asking a Snapchat lens, "Hey, how far is Neptune?" and boom, you get an answer right then and there. They're also talking about using this tech for quizzes or random fun facts.

Snap's already got a ChatGPT thing going on with My AI, a chatbot that's like texting with a buddy. First, it was a special deal for the Snapchat Plus crowd, but now everyone can use it for free. They even tried putting ads in the chatbot's replies. But, heads up, there's some heat from a UK data rights group about user privacy.

And for the tech-savvy, Snapchat's dishing out new tools for lens makers. They're bringing in a 3D face mask generator and making it easier for teams to collaborate on filter projects. Looks like the future of filters is gonna be pretty wild with AI in the mix!

READ THE ARTICLE ON THE VERGE

Microsoft briefly restricted employee access to OpenAI’s ChatGPT, citing security concerns

Microsoft, the big cheese in the tech world, has poured a ton of cash into OpenAI. But for a hot minute on Thursday, their folks couldn't use ChatGPT, OpenAI's star product. Turns out, Microsoft got antsy about security and data risks, so they hit pause on employees using ChatGPT and some other AI tools.

They didn't say much when asked, but they did mention that even though ChatGPT has some safety nets to stop it from being misused, it's still an outside service, so you gotta be careful. This caution applies to other external AI services too.

Microsoft's advice? Use their own Bing Chat, which also runs on OpenAI's smarts. These two are thick as thieves. Microsoft's been updating its Windows and Office stuff to use OpenAI's tech, which runs on their Azure cloud.

READ THE ARTICLE ON CNBC

Humane officially launches the AI Pin, its OpenAI-powered wearable

Humane just showed off their new gizmo, the AI Pin. It's a $699 wearable that clips onto your clothes with a magnetic battery pack. Plus, you gotta shell out 24 bucks a month for a subscription, which gives you a phone number and data from T-Mobile. They're gonna start shipping it early 2024, and you can order one starting November 16th.

This thing runs on a Snapdragon processor (not sure which one, though) and you use your voice, gestures, and a little projector to control it. The Pin weighs about an ounce, and the battery adds a bit more. It's got a camera that snaps 13-megapixel photos and will do video later on with an update.

The AI Pin's big thing is connecting to AI models, like ChatGPT, without needing a bunch of apps. Humane's trying to make tech less complicated – no home screens or tons of settings. Just talk or touch the Pin, and it does what you need.

But what can it really do? Well, it can send messages, make calls with your voice, summarize your emails, figure out food's nutrition from a picture, and translate languages in real time. They're planning to add navigation and shopping, and let developers make their own tools for it too.

READ THE ARTICLE ON THE VERGE

Google wants AI chatbot Bard to help it reach billions of users

Google's cooking up this new chatbot called Bard. The big idea? Make something as huge as their search engine or YouTube. They're hoping it'll be a game changer, getting more folks to use their stuff.

Jack Krawczyk, the head honcho for this project, spilled the beans at this Reuters shindig in New York. He's jazzed about mixing Bard with Google Assistant. Think of it like giving Assistant a boost with some smart, human-like tips. They're rolling this out on phones first, aiming to get more people on board with AI.

Google's parent company, Alphabet, has a bunch of products with a bazillion users. Now they're betting big on AI to keep up with the Joneses. Amazon's doing something similar with Alexa, and OpenAI's already jazzed up ChatGPT with voice stuff.

But here's the thing - how we find info is changing. Bard's getting some attention, but not as much as ChatGPT. Even Google Search took a slight hit. Google's more focused on making Bard useful than figuring out how to make dough off it, like with ads or subscriptions.

READ THE ARTICLE ON REUTERS

OCBC’s new generative AI chatbot is boosting the bank’s productivity across departments and locations

OCBC Bank in Singapore is using a new AI chatbot to make their work easier and more efficient. This chatbot, which employees like Denise Law use, helps figure out what customers want from their long emails, like checking account balances or getting fee waivers. The chatbot can even make responses sound less robotic.

This tool isn't just for emails. It also helps update frequently asked questions, translate words, and even draft replies for customer complaints. It's super fast and makes sure customers get consistent answers, which improves service big time.

OCBC, Singapore's second-largest bank, got into this AI chatbot game after seeing how cool ChatGPT was. They built it in a secure way on Microsoft Azure and added it to Microsoft Teams for easy access. The bank's folks have found all sorts of neat ways to use it, like front office managers doing industry research, call center staff summarizing calls, and HR folks writing job descriptions.

READ THE ARTICLE ON MICROSOFT

NVIDIA Turbocharges Generative AI Training in MLPerf Benchmarks

NVIDIA just knocked it out of the park in the latest MLPerf benchmarks, showing off their AI muscle. Their new beast, the NVIDIA Eos, is a powerhouse with over 10,000 GPUs. It crushed a training benchmark for a GPT-3 model, like the one behind ChatGPT, in just under 4 minutes. That's almost three times faster than their last record.

This speed means businesses can train massive language models way faster, cheaper, and with less energy. NVIDIA's got this tool, NeMo, that lets companies tweak these big brain models for their own needs.

In another test, they used their Hopper GPUs to train a text-to-image model, like the one behind those cool AI-generated pictures, in just 2.5 minutes. These tests are super important because they're like the gold standard for measuring AI performance, and generative AI is the big thing right now.

Tons of GPUs and smart software tweaks. NVIDIA's scaling up big time, showing they can handle the huge demands of today's data centers. Both their Eos system and Microsoft Azure used over 10,000 GPUs and showed almost identical top-notch performance.

READ THE ARTICLE ON NVIDIA

Eleos Health raises $40 million to extend the use of AI for behavioral health

Eleos Health, a company from Israel that kicked off in 2020, just bagged a cool $40 million in its latest cash round. This boosts their total dough to $68 million. They're all about using smart tech to make mental health care better. The big dogs leading this funding spree are Menlo Ventures, plus some other heavy hitters like F-Prime Capital and SamsungNEXT.

The trio who started this gig – Alon Joffe, Alon Rabinovich, and Dror Zaide – got a boost when Dr. Shiri Sade Sharvit, a brain doctor, joined them. Their big thing? VoiceAI, which is like having a smart assistant in therapy sessions, picking up on all sorts of details to help therapists do their thing better.

Eleos is on a roll, employing around 95 folks worldwide, and they've ramped up their team by over half since last year. They're also big on gender balance, with women making up about half the team and 40% of their bosses. They're planning to hire 50+ more people by the end of 2024.

READ THE ARTICLE ON CTECH

Public AI Training Datasets Are Rife With Licensing Errors 

A bunch of smart folks, including some from MIT and a nonprofit called Cohere for AI, found a big problem with the data used to train AI, like those language models we chat with. They checked out over 1,800 datasets from places like GitHub and found that most of them didn't have clear legal rights or licenses. This is a big deal because it means people using this data might be stepping on legal toes without knowing it.

A lot of the datasets didn't have any license info, and when they did, about half were wrong. Turns out, nearly 30% were more chill than the creators meant them to be. This could mean trouble for folks trying to play by the rules.

There's a whole mess with how this data is collected too. A lot of it is just yanked from the web, and that's got some writers and artists pretty steamed, saying it's like stealing their stuff. The Atlantic even made a tool to let authors check if their work's been nabbed for AI training.

The big issue here is that lots of these datasets are mishmashes of smaller ones, each with different rules. This makes it super confusing to figure out if you're using the data right. The team behind this research made a tool to help sort this out, but the whole situation shows that folks in AI aren't taking the legal side of data use seriously enough. There's a lot of gray area in how the law looks at this, and it's tough to track if someone's breaking the rules.

READ THE ARTICLE ON IEEE SPECTRUM

AI maps icebergs 10,000 times faster than humans

A team of brainy folks at the University of Leeds cooked up an AI that's a real game-changer for tracking big icebergs in Antarctica. This neural network is a whiz at spotting icebergs in satellite snaps, doing the job in a blink – just 0.01 seconds! Before, this was a slow grind, done by hand.

Anne Braakmann-Folgmann, the lead researcher, now at the Arctic University of Norway, points out how crucial these giant icebergs are. They mess with the ocean's physics, chemistry, and biology and are a big deal for ships cruising around there.

The Copernicus Sentinel-1 radar mission is key to this AI trick. It gives us the lowdown on icebergs even when it's cloudy or dark. Regular satellite photos make it tough to tell icebergs from sea ice and clouds because they all look white. But with radar images, icebergs pop out bright against the darker sea and ice.

They based this neural network on a fancy U-net design, trained with a bunch of Sentinel-1 shots showing huge icebergs. The AI got better by comparing its guesses to the real deal until it was top-notch.

They tested this on seven icebergs, ranging from the size of Bern, Switzerland, to Hong Kong. The dataset covered different seasons from 2014 to 2020, with one photo per month for each iceberg. And guess what? This AI nailed it with 99% accuracy.

READ THE ARTICLE ON THE EUROPEAN SPACE AGENCY

Sam Altman just roasted Elon Musk's new 'Grok' AI chatbot, calling it 'cringey boomer humor'

Sam Altman, the big cheese at OpenAI, just threw some shade at Elon Musk's new AI toy, "Grok". Grok is like a new kid on the block trying to square up against ChatGPT, but with a twist – it's all about making jokes. But not just any jokes – we're talking "cringey boomer humor".

Here's the backstory: Altman and Musk used to be tight. They were part of the crew that kicked off OpenAI back in 2015. But things got rocky when OpenAI, under Altman's lead, decided to buddy up with Microsoft and took a step back from being just a non-profit.

So, Altman's got his own thing going with these custom ChatGPTs you can tweak to your liking. And he goes ahead and sets one up to mimic this Grok thing, making it spit out awkward, try-hard jokes. That's a pretty bold move, especially right before OpenAI's big DevDay event where they're showing off these fancy custom ChatGPTs.

READ THE ARTICLE ON BUSINESS INSIDER