AI News That Matters - July 22nd, 2024

Meta to launch biggest Llama 3 model, NVIDIA unveils new Chinese chip, OpenAI avoids another lawsuit and more!

Author

Everyday AI
July 22, 2024

Outsmart The Future

Sup y’all 👋

Clear your calendars – you’re REALLY gonna want to join us LIVE this Tuesday at 7:30 AM CST. As we prepare to launch our ‘Thanks a Million’ campaign.

We’re celebrating by giving away $1,000! (If anyone can ace our test, that is)

Yep. You read that right.

Join us live to compete for your chance to win as we dish out questions about ChatGPT and other LLMs!

If you get ALL questions correct in real-time, you could walk away with $1,000.

✌️

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: From OpenAI’s new GPT-4o mini to big tech stealing YouTube content, here’s what you need to know for this week’s AI news that matters! Give it a listen.

🕵️‍♂️ Fresh Finds: Yandex co-founder launches AI startup, the CIA’s AI plans and the current state of AI adoption. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Meta to release largest Llama 3 model, NVIDIA to produce new Chinese AI chip and OpenAI avoids another lawsuit. For that and more, read on for Byte Sized News.

🚀 AI In 5: Is GPT-4o actually better than GPT-4 Turbo? We’re showing you an easy way to compare the two. See it here

🧠 AI News That Matters: Need to catch up on the latest AI news? We’ve got you covered. From GPT-4o Mini to tech giants stealing YouTube content, here’s what you need to know. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about GPT-4o Mini, OpenAI wanting to make an AI chip, big tech uniting for AI security and Mistral and NVIDIA creating an LLM. Check it here!

AI News That Matters - July 22nd, 2024 📰

A new large language model from the industry leader.

Huge updates in AI lawsuits.

International turmoil around AI regulation.

That's just the beginning. 

This week was a chaotic one in AI news. What's it all mean for your biz?

We got you covered with this week’s AI news that matters.

Join the conversation and ask Jordan any questions on AI here.

Also on the pod today:

• Release of OpenAI's GPT-4o Mini 🤖
• Use of Copyrighted Content to Train AI ❌
• Launch of AI Education Platform, Eureka Labs 🧑‍🏫️

It’ll be worth your 47 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Upcoming Everyday AI Livestreams

Tuesday, July 23rd at 7:30 am CST ⬇️

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Vozo turns videos into stories and prompts, Gobi is an AI health guide and Studymap.ai helps you learn anything.

Trending in AI - Yandex’s co-founder and former CEO has launched an AI startup in hopes to become a EU AI compute powerhouse.

Money in AI – AI startup Cohere has raised $500M from new investors.

AI in Politics – The CIA’s AI director spoke on its strategic AI approach.

AI in Society - A county in Michigan is using AI videos to help victims of crime.

Read This – A recent survey shows the current state of AI adoption in companies worldwide.

1. Meta To Release Largest Llama 3: The 405 Billion Parameter Behemoth 🦙

Meta Platforms is set to release its largest Llama 3 model tomorrow, July 23, boasting an impressive 405 billion parameters. This multimodal model will not only comprehend and generate text but will also have the capability to create images, significantly enhancing its functionality.

Following the successful launch of smaller models with 8 billion and 70 billion parameters earlier this year, the anticipation for this major release has been building since it comes just a year after Llama 2

2. NVIDIA to Start Mass Production of AI Chip for China 🇨🇳

NVIDIA is stirring the pot by preparing to launch a new flagship AI chip, dubbed the B20, specifically designed for the Chinese market while following U.S. export controls. The company’s Blackwell chip series, expected to be 30 times faster than its predecessor, is set for mass production later this year, with hopes to ship the B20 by Q2 2025.

NVIDIA's shares skyrocketed by 14% to $119.67 after this news broke, proving that even amidst U.S. sanctions, the demand for advanced tech in China is hotter than ever.

3. OpenAI Dodges a Bullet in Copyright Lawsuit 🧑‍⚖️

OpenAI has escaped a copyright lawsuit from a group of open-source programmers who claimed their code was used to train its AI without giving them proper credit. The programmers decided to voluntarily dismiss their case against OpenAI but are still gunning for GitHub and Microsoft, who teamed up with OpenAI on this project.

While the judge tossed out claims related to the Digital Millennium Copyright Act, the open-source licensing allegations are still very much on the table

4. LinkedIn Adds AI to Boost Daily Visits 🚀

LinkedIn is aiming to get users to check in daily with its new AI career advice feature. The platform is redefining itself to compete with the entertainment kings like TikTok and Facebook, noting a decline in user visits, with only a 5% growth in site visits year-over-year.

Chief Editor Daniel Roth says they're on a mission to make LinkedIn a daily habit, and they're also eyeing younger shoppers, with Gen Z leading the way in social media shopping.

5. Google's New AI Weather Model Could Change Forecasting 🌦

Google researchers have developed a groundbreaking climate model called NeuralGCM that merges artificial intelligence with physics, promising more accurate weather predictions while significantly reducing computer power requirements.

Unlike traditional models that rely on supercomputers and extensive processing time, this innovative tool can generate forecasts in just 24 hours using a single chip.

6. Capgemini Predicts Autonomous AI by 2025 👀

Capgemini has unveiled a significant trend in their latest report, showing that 82% of companies are planning to integrate AI agents within the next three years. These advanced agents are designed to operate independently, potentially taking over tasks traditionally performed by humans.

Chief Innovation Officer Pascal Brier emphasized the emergence of multi-agent AI systems, where these agents collaborate seamlessly to enhance productivity.

7. Elon Musk's xAI Unveils Most Powerful AI Training Cluster 💪

Elon Musk's xAI has debuted the Memphis Supercluster, now the globe’s most potent AI training system, featuring 100,000 liquid-cooled H100s. According to Musk's announcement on his X account, this cutting-edge cluster, born from a collaboration with Nvidia, promises to supercharge AI development.

Located in Memphis, Tennessee, it's poised to be a pivotal hub for advanced AI research. This milestone signifies a leap forward in AI capabilities, highlighting how strategic partnerships can drive innovation.

ChatGPT Showdown: How to evaluate GPT-4o vs. GPT-4 Turbo

Some people are claiming that GPT-4o has taken a step down in quality.

Is that true?

We’re showing you a simple way to compare GPT-4o and GPT-4 Turbo. So then, you can judge for yourself.

Trying to find the Top 5 AI news stories this week was harder than finding a parking spot at the mall on Black Friday. 

How did OpenAI reportedly making its own chips or Anthropic’s $100 million AI startup fund not make our Top 5 roundup? 

Well…. This week was A LOT. 

Tech giants caught red-handed, there’s legal drama that could rewrite AI's rulebook, and a familiar face bringing AI education to the masses. 

Let's break it down shorties. 👇

1 – YouTube's closed captions: AI's new favorite snack? 🎥

Some tech bigwigs are reportedly in hot water for using YouTube subtitles to train their AI without permission. 

We're talking 170,000+ videos from 48,000 channels, including big names like Mr. Beast and MKBHD, as well as mainstay media companies like ABC News and The New York Times. 

The culprits?

This data's part of a bigger collection called "The Pile," which includes books and Wikipedia articles too.

MKBHD (aka Marques Brownlee) called it an "evolving problem." Understatement of the year, Marques.

What it means: 

This clash between AI development and content rights could lead to stricter regulations on data usage for AI training. 

For businesses, it's a wake-up call to scrutinize the origins of their AI training data. The legal landscape is shifting, and you don't want to be caught off guard.

But, we’ll cut it to you straight, y’all.

The majority of content LLMs are trained on is actually copyrighted data. How will that all shake out? 

#2 below might ultimately have the answers. 

2 – OpenAI to NY Times: Prove it or lose it 🥊️

OpenAI's reportedly demanding the New York Times provide evidence for millions of articles in their ongoing legal battle.

The Times accused OpenAI of using their content without permission to train ChatGPT. 

OpenAI's counter? The Times manipulated their AI to reproduce content.

This massive request could strain the Times' resources and ability to pursue the lawsuit. It's a bold move that underscores the complexity of AI and copyright law.

What it means:

The outcome could set major precedents for how AI companies interact with content creators. 

In theory, this case could reshape copyright law in the digital age, impacting how companies use and interact with AI technologies.

We think this case is too big to be litigated, and it’ll likely be settled out of court. And like OpenAI, we think the NYT and their legal team straight fumbled the bag here.

Check out this episode for a WAYYYY deep dive. 

3 – OpenAI cofounder starts Eureka Labs in big education bet️ 🧑‍🏫️

Andrej Karpathy, former OpenAI co-founder and former Tesla AI head, is diving into education with Eureka Labs.

The startup aims to create AI teaching assistants to guide students through course materials.

Their first offering, LLM 101N, will reportedly help students build their own storyteller AI using Python, C, and CUDA.

Karpathy hints at future AI assistants based on real people, similar to Meta's celebrity chatbots.

What it means:

Most of the times when super smart AI masterminds leave a company, it’s to start their own OpenAI competitor. 

Karpathy is taking a different path, and we’re here for it.

We’ve been saying since Day 1 how education is such an overlooked piece of the Business AI puzzle. 

The smartest people in the world are building AI, and hardly no one is properly teaching it and helping companies cover their bases. 

4 – Meta gives EU the cold shoulder on AI 🥶

Meta's withholding its new multimodal AI model from the EU due to regulatory concerns, according to reports from Axios and others.

Meta’s updated Llama model will be able to handle video, audio, images, and text, but won't be available in Europe.

(Side note — keep an eye out on tomorrow’s newsletter, as Meta is set to release its updated Llama 3 model.) 

Apple's made a similar move, keeping some AI features out of EU iPhones.

It's part of a growing trend of U.S. tech giants clashing with European regulators after the EU passed the EU AI Act, which goes into effect August 2026. 

What it means: 

Things are gonna get wild in the EU. 

The EU’s AI Act is the first major multinational legislation on Artificial Intelligence.

While we applaud the EU for taking these measures, we think this is one of those instances where being first isn’t necessarily a good thing. 

(But someone’s gotta do it, right?) 

The EU’s AI Act could create a "tech gap" between the EU and other regions.

For businesses with international operations, navigating these regional differences will be crucial.

It also raises questions about the balance between regulation and innovation in the AI space.

5 – OpenAI's new mini-me: GPT-4.0 Mini 🌟

We gotta save the banger for last. 

OpenAI just launched GPT-4o Mini, a smaller, cheaper version of their powerhouse LLM, GPT-4o. 

First things first — this is mainly a developer play.

If you’re a “normal” front-end ChatGPT user paying $20/mo for ChatGPT Plus, you’ll still want to use the GTP-4o model for most use-cases. 

But if your company has its own internal model, or if you’re a startup building a product or service on top of the GPT tech, GPT-4o Mini will legit change the game. 

It's 60% cheaper than GPT-3.5 Turbo and 90% cheaper than the original GPT-3.5.

Despite its size, it's outscoring competitors like Google’s Gemini 1.5 Flash and Anthropic’s Claude 3 Haiku on key benchmarks. 

This model is aimed at developers and businesses looking for cost-effective AI solutions.

What it means: 

People are looking at this Mini model as just a dummied down version of GPT-4o.

Naaaaaaah shorties. You’ve got it all wrong. 

This changes the future of AI in business. As an example, two years ago it woulda been a bad investment for most companies to fine-tune their own models. 

Now, it’s almost a given. 

If you’ve got someone with the tech know-how on staff, you HAVE to be marrying your data with your LLM of choice. 

And this isn’t just a ‘choose OpenAI because it’s the best’ segment. 

And with GPT-4o Mini (right now) wiping the floor against other Small Large Language Models, the AI future is exciting. 

And powerful. 

And cheap. 

⌚

Numbers to watch

81%

According to a survey, 81% of companies plan to increase AI spending next year.

Now This …

What are your thoughts?

Vote to see live results

If you don't listen to our podcast, why not?

Login or Subscribe to participate in polls.

Reply

or to participate.