Today:

OpenAI’s app store for GPTs will launch next week

OpenAI is about to launch its own GPT App Store next week. This new store lets developers showcase their custom AI apps, built using OpenAI's text-gen models like GPT-4. Before listing, devs gotta check out OpenAI's new rules and make sure their stuff fits the bill. Last year, OpenAI hinted at this during their DevDay but had to push it back, probably because of some big changes at the top. 

Making these apps is easy-peasy, no coding needed. You just tell the GPT Builder what you want, and you got an AI chatbot. Up till now, devs could share their GPTs but couldn't list them publicly. Big question still hanging is how the money stuff will work in this store. OpenAI's keeping that under wraps till the launch. 

TECHCRUNCH

Google appears to be working on an ‘advanced’ version of Bard that you have to pay for

Google's cooking up something cool with "Bard Advanced." It's a fancier version of their Bard chatbot, but you gotta pay to use it. This new Bard's brainier, using Google's latest tech, Gemini Ultra, for smarter answers and tricky math. 

A dude named Dylan Roussel found this out and shared some sneak peeks. There's also talk about making your own custom bots with a thing called Motoko, but it's not clear if that's free. Plus, Google might add a "power up" feature to make your Bard chats even sharper and a "Gallery" to explore different topics. They're also adding a tab for keeping track of bigger tasks like making images. No word yet on when this stuff goes live, but Google's been busy beefing up Bard to keep up with ChatGPT and others.

THE VERGE

Introducing a new Copilot key to kick off the year of AI-powered Windows PCs

Microsoft's introducing a game-changer for Windows 11 PCs: the Copilot key. It's a big deal, the first major keyboard update in almost 30 years. This key makes using AI on your computer super easy, just like the Windows key did back in the day. It's all about making computers smarter and more user-friendly, with AI built right into the system.

They've been working hard on this, making AI a daily helper with stuff like Bing search and Microsoft 365. And now, with partners like AMD, Intel, and Qualcomm, they're really kicking it up a notch. This key is a big step in their journey to make AI a regular part of our computer life. You'll start seeing it on new Windows 11 PCs, including Surface devices, from now through spring. It's a bold move into a future where AI and PCs are best buddies.

MICROSOFT

Shaping the future of advanced robotics

The Google DeepMind Robotics Team just dropped some serious tech on us with AutoRT, SARA-RT, and RT-Trajectory. They're making robots that get stuff done like cleaning your house or whipping up a meal. AutoRT uses big brain models to train robots better with real-world tasks. 

They tested this with up to 52 robots, doing over 77,000 trials in offices. Safety's a big deal here, so they've got rules like Asimov's "robots can't hurt humans" and other no-nos like sharp objects. They've also got physical kill switches just in case.

SARA-RT is about making robots think faster without losing smarts. RT-Trajectory is for teaching robots how to move. It uses videos with overlaid sketches to show robots how to do tasks, like wiping a table. 

GOOGLE DEEPMIND

Search startup Perplexity AI valued at $520 mln in funding from Bezos, Nvidia

Perplexity AI, a search startup, just bagged $73.6 million in funding, shooting its value up to $520 million. Big names like Nvidia and Amazon's Jeff Bezos are in on it. They're betting big on AI, and Perplexity's right in the thick of it. Their tech? It's all about quick, smart answers using various big-brain language models from OpenAI to Meta's Llama. 

Based in San Francisco, these guys have been busy, handling over 500 million queries last year with minimal marketing. They're small but growing fast, with plans to boost their team to 60 by year's end. Their online traffic's exploded, from 2.2 million to 45 million hits in just a month!

REUTERS

🧠RESEARCH

Improving Text Embeddings with Large Language Models

Microsoft introduces a new, simpler way to create high-quality text embeddings, avoiding the usual complex and big steps. They use fake data and less than 1,000 training steps, unlike other methods that need many stages of pre-training and fine-tuning with lots of text pairs and a few labeled datasets. Tests show their method works great without any real labeled data. Plus, mixing fake and real data, their model reaches top-notch results on major text embedding benchmarks.

What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs

A team from NVIDIA and UC San Diego developed a new method to make 3D images look super real using fewer computer resources. Usually, creating these images takes a lot of power and memory, but these folks figured out a way to do it better and faster, keeping the 3D look sharp and consistent from all angles. Their technique is like a smart shortcut that cuts down on the heavy lifting but still gets top-notch results. This is a big deal for making 3D images more realistic without needing a supercomputer. They tested it on some standard 3D image sets and set new records for quality.

ODIN: A Single Model for 2D and 3D Perception

"ODIN: A Single Model for 2D and 3D Perception" is a game-changer in the tech world. Created by hotshots from Carnegie Mellon, Stanford, and Microsoft, it's a model that can handle both flat pictures and 3D scenes. Until now, people thought you needed different tools for 2D and 3D stuff, but ODIN proves them wrong. It uses a smart design that switches between 2D and 3D, keeping track of where things are in space. It's nailing tests on various benchmarks, beating out older methods, especially when it gets the raw 3D data. Plus, it's rocking it in a setup where it has to understand instructions and act on them. 

🛠️TOP TOOLS

AI Fitness Planner - offers custom workout plans, personalized meal suggestions, and body composition analysis, all powered by AI, to help you achieve your fitness goals efficiently. 

RecordOnce - lets you make professional video tutorials in minutes. It uses AI for editing and fixing mistakes, making video creation easy for anyone. It records interactions instead of pixels, offering advanced editing options and even automatic voice-overs, all aimed at streamlining the video creation process.

Rawuser - designed to enhance website user engagement. It provides tailored user experiences and content, dynamically improving user interaction and conversion rates. 

Symbolab - comprehensive math solution tool offering step-by-step explanations for various math problems. It includes features for graphing, geometry, and a wide range of calculators. 

AImReply - streamlines email writing and response. It allows users to efficiently create personalized, error-free emails in multiple languages. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Elvis show to debut in London featuring AI-generated life-size projection of iconic artist

Elvis Presley's hitting London with a high-tech twist! A show featuring a life-size, AI-generated Elvis avatar's gonna debut in London, then head to Vegas, Tokyo, and Berlin. It's a mix of AI, augmented reality, and loads of cool effects, telling Elvis' story and rocking a concert vibe. This tech magic, already a hit with ABBA's show in London, is sparking chatter about ethics in digitally resurrecting artists. CNBC

AI prescription assistant nabs backing from Mayo Clinic, Eli Lilly

A startup, UpDoc, launched a voice-based AI app for managing prescriptions like insulin. Supported by big names like Mayo Clinic and Eli Lilly, it's part of the booming AI device market. Created by Stanford docs, the app helps monitor and intervene for patients at home, especially those with chronic conditions. It's clinician-guided and shows promise in improving chronic care, notably in a study with Type 2 diabetes patients. The name and wake word, inspired by Bugs Bunny, add a quirky touch. Venture capital and tech giants are also backing it. AXIOS

NIST Identifies Types of Cyberattacks That Manipulate Behavior of AI Systems

NIST warns AI systems can be tricked or "poisoned" by bad guys, messing them up without a surefire fix. Their new report, part of a bigger project for reliable AI, talks about these risks and how to handle them, but admits there's no perfect solution. They point out how AI, used in stuff like self-driving cars or chatbots, learns from huge data sets that can be tampered with, leading to screw-ups like offensive chatbot replies. The report digs into four attack types—evasion, poisoning, privacy, and abuse—and suggests ways to lessen the damage, but there's still no rock-solid defense. NIST

US regulator denies Apple, Disney bids to skip votes on AI

The U.S. Securities and Exchange Commission ruled that Apple and Disney can't dodge shareholder votes on their AI use, as demanded by a major labor group. Despite the companies' argument that this involves "ordinary business operations," the SEC sees it as a significant issue, especially given recent concerns about AI replacing jobs and misusing creative work. The AFL-CIO, America's largest labor union, is pushing for more transparency and ethical guidelines in AI use by these companies. REUTERS

ChatGPT could soon replace Google Assistant on your Android phone

ChatGPT might soon be as easy to use as Google Assistant on Android phones. OpenAI's ChatGPT app code hints at a future update where you can set it as your default assistant. This means quick access without opening the app every time. But there's a catch: it's not fully ready yet. The app's missing key code pieces for seamless integration. Plus, it can't create or respond to custom hotwords like Google Assistant. OpenAI is also adding a Quick Settings tile for faster access, but it's not working yet. ANDROID AUTHORITY

AI agents help explain other AI systems

MIT researchers developed a new method where AI explains AI, using "automated interpretability agents" (AIAs). These AIAs, built from pretrained language models, actively experiment on other AI systems to understand and explain their behavior. This approach is a leap from traditional passive methods, offering deeper insights into complex AI models. Additionally, they introduced the "function interpretation and description" (FIND) benchmark, providing a standard for assessing AI explanations and advancing AI interpretability research. MIT

How Google is using AI to help one U.S. city reduce traffic and emissions

Google's teaming up with a U.S. city to cut down on traffic jams and pollution. They're tweaking traffic lights using AI and their massive map data. Even a few seconds' change can ease traffic and reduce emissions. Seattle's testing it, and Google aims to roll it out in more cities. This could mean smoother drives and cleaner air, a big win for both drivers and the environment. CBS NEWS

Create Custom GPTs | No code AI automation | OpenAI App Store | Complete Guide

The GPT Builder is a new tool that lets users create customized GPTs to automate various aspects of life and business. Here’s a complete guide of building different GPTs, like an expense tracker and a tool to analyze company 10K reports. This technology could significantly streamline and enhance tasks, especially for businesses.