Llama 3 & Meta AI, Earbuds with ChatGPT, AI Pin Failure, and Model Turns Images Into Talking Faces
PLUS HOT AI Tools & Tutorials
Welcome to a breakdown of the last week's AI news!
Today, we’ll discuss news for all tastes: impressive announcements from Meta, disappointing results from AI Pin, and frightening advances from Microsoft researchers. And that's not all. Following them, we will mention Sam Altman's ongoing world tour, the first earbuds with ChatGPT, and other important events.
Let's get started!
This Creators’ AI Edition:
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
Do you have a minute to participate in Subscription Giveaway and make our newsletter even better? Take a short survey and tell us about your experience + participate in the free Giveaway of a Premium Subscription for 6 months.
Featured Material 🎟️
Meta Releases New Llama 3
The battle to be the best AI developer in Silicon Valley continues. This week, Meta has made many big AI-related announcements. But they're all related to releasing two new models in its Llama 3 family: the Llama 3 8B and the Llama 70B. That's why we should start with him—briefly, without boring figures (almost) and only the most important things.
As the name suggests, the Llama 3 8B model has 8 billion parameters, while the 70B has 70 billion. The developers say they have made a "major leap" from previous models and call this family among the best open models available.
According to popular benchmarks (shown in the picture below), Llama 3 outperforms Google and Mistral models. Besides, according to Meta's benchmarks (well, who would doubt it), the new models cope better with real tasks, like coding or writing, than Mistral Medium from Mistral, GPT-3.5 from OpenAI, and Claude Sonnet.
Meta says that users of the new Llama models should expect more “steerability,” a lower likelihood of refusing to answer questions, higher accuracy on trivia questions, questions about history and STEM fields such as engineering and science, and general coding recommendations. These results are due to the large dataset.
The developers used a collection of 15 trillion tokens and about 750 billion words, seven times larger than the Llama 2 training set. That sounds impressive.
The Llama 3 family is already available on Meta's official website. And you can learn more about Llama 3 tests and their performance from its press release.
Keep your mailbox updated with key knowledge & news from the AI industry
Llama 3 with Meta AI
The release of the Llama 3 is only part of the story. Along with the release of the new models, Meta announced that the Meta AI assistant, introduced last September, will now be available in the Instagram, Facebook, WhatsApp, and Messenger search boxes. And it has also gotten a separate website, Meta.ai.
Looks like Meta is ready to compete directly with ChatGPT and is attacking on all fronts!
Like other chatbots, Meta AI answers user questions, dialogs with natural language, and helps users search for information. But this model has one big feature. This AI can aggregate search results from both Google and Bing and answer user queries.
It is the only model capable of doing such a job.
Image Generation in WhatsApp
Meta has also launched a beta version of AI-assisted image generation for WhatsApp users. The feature is currently available for a limited number of users in the US but will soon be available in other countries.
Of course, like other companies, Meta calls its models and chatbots the best and most productive. But real-world results often differ from dry tests. In any case, we'll soon find out what Meta's new AI can do.
By the way, Mark Zuckerberg's company won’t stop with this update. Meta said it will release several more models with additional capabilities in the coming months. One of them is an even larger model with 400B parameters!
Sharing is caring! Refer someone who started a learning Journey in AI!
News Of The Week 🌍
Humane AI Pin is Now Available… And it’s failure
A few weeks ago, we reviewed AI gadgets. Among them was the AI Pin, a small square brooch that promised to replace your smartphone with algorithms partially. At the time, the device seemed very promising. And. if you don't remember, developers from Humane said AI Pin "will reimagine the human-technology relationship."
But it turned out to be quite different.
According to The Verge, almost all major reviewers agreed that no one should buy this gadget. AI Pin is very slow, its operating system is inconvenient, and the announced functions don’t work. In particular, the gadget cannot translate, set an alarm or timer, or use a calendar. Sad, but we have what we have.
You can check out Marques Brownlee's review (the usual very optimistic one), which pretty accurately shows “the real usefulness” of AI Pin:
In case you missed it, here's our review of several gadgets, including the AI Pin. Then, the device from Humane looked promising, but something seems to have gone wrong:
Google Consolidates DeepMind and Research Teams
Google has figured out how to speed up the development of new AI models. The answer, in general, is obvious: the two leading teams in this field, Research and DeepMind, should be merged. According to Reuters, it will relocate Responsible AI teams—which focus on safe AI development—from Research to DeepMind so that they are closer to where AI models are built and scaled.
DeepMind was a team that helped develop Google's most advanced model, Gemini, unveiled late last year. We made a detailed breakdown of this chatbot:
Microsoft New Model Turns Images Into Talking Faces
Microsoft knows how to scare us. The company's researchers showed VASA-1, a new model that can create a deepfake in seconds. It doesn't require a text description or complex settings - the AI just needs a single image of a person and an audio track. Based on this data, VASA-1 will generate a video of a talking person with emotions and natural (well, almost) facial expressions.
Here's what it looks like:
You can also see more examples on Microsoft's official site.
Nothing Ear Introduced New Wireless Earbuds
British startup Nothing has unveiled two new models of its earbuds lineup. It would seem that this has nothing to do with our newsletter. (Heck, you saw the reason at the very beginning.) Yes, these are the first earbuds with native ChatGPT support.
I won't get into specs and tech details, but I’ll just say that my Nothing Ear (1) makes me happy every day.
Soon, owners of the new Nothing Ear paired (unfortunately) with the company's smartphones will be able to use OpenAI's chatbot as their primary voice assistant. The principle will be identical to Siri on iPhones or Google Assistant on Android. Users will not need additional customization; everything they need is built into the Nothing app.
That sounds great, but after the AI Pin results, I want to see real-world examples.
OpenAI Opens First Office in Tokyo
It seems that it is time to open the column "Around the World with Sam Altman".
After Altman’s trip to South Korea and UAE, OpenAI announced the opening of its first office in Japan. According to the CEO, this is just the first step in a long-term partnership with the people of Japan. The company also said it has a custom model optimized for the Japanese language and that Tadao Nagasaki, ex-president of Amazon Web Services in Japan, is heading the local business.
OpenAI's customers in Japan include automaker Toyota Motor, manufacturer Daikin Industries, and local government.
Creators’ AI Subscription is the best gift for Creators diving into AI.
Useful Tools ⚒️
Tabnine – The private, personalized, & protected AI coding assistant
Brave Search – AI-powered answer engine
Tabula – Startup analytics, accelerated by AI
Collato AI Notetaker – A smart notetaker using machine learning
Collectif AI is an AI-powered software platform specifically designed to help SaaS (Software as a Service) companies streamline their user research and extract actionable insights from vast amounts of customer data. It connects to existing tools like support ticketing systems, feedback platforms, and sales call recording software, creating a centralized hub for all your customer data.
Weekly Guides 📕
Udio Tutorial - How To Make A Full Song - Detailed Guide
Save HOURS "Researching" in ChatGPT! (Full Guide)
Use AI to Swap Lil Yachty with Joker in the LEGENDARY Walk-out Video
Build a Retrieval Augmented Generation (RAG) AI
AI Meme Of The Week 🤡
AI Tweet Of The Week
A few days ago, Sam Altman (along with OpenAI's COO Brad Lightcap) participated in a podcast for the YouTube channel 20VC. There, he gave some details about ChatGPT and talked a lot about startups and AI and startups.
Follow this link if you missed it!
(Bonus) Material 🏆
How One Author Pushed the Limits of AI Copyright
The Smartest Way to Use AI at Work
U.S. tech companies dominate the generative AI boom (and here’s why)
Types of Artificial Intelligence That You Should Know in 2024
Take our survey to participate in the Premium Subscription Giveaway!