• Everyday AI
  • Posts
  • 5 Things to Know about Meta's Llama 3.1

5 Things to Know about Meta's Llama 3.1

Salesforce's huge AI shift, NVIDIA and Microsoft being sued, global tech titans unite for LLM security and more!

Author

Everyday AI
September 06, 2024

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: This summer, Meta released a bombshell with THREE new AI models. Does OpenAI finally have some serious competition? If you haven’t had time to kick around Llama 3.1, we break down the 5 things you need to know. Give it a listen.

🕵️‍♂️ Fresh Finds: An OpenAI investor speaks on Elon Musk, a tuned Meta model is throwing around some major weight, DeepMind unveils new protein AI model and Salesforce acquires AI startup. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Salesforce introduces AI agents, a startup sues NVIDIA and Microsoft and global tech titans unite for LLM security. For that and more, read on for Byte Sized News.

🚀 AI In 5: Kling AI text to video has been released globally. Does it stand its ground against Runway? Let’s find out. See it here

🧠 Learn & Leveraging AI: Here’s what you need to know about Meta’s new AI models and how they impact the AI landscape. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about OpenAI considering price changes, a new international AI treaty and why your business shouldn't use Claude. Check it here!

5 Things to Know about Meta's Llama 3.1 🦙

Another one!?

Another day, another HUGELY impactful model.

Days after OpenAI dropped the game-changing GPT-4o Mini, Meta responded with a one-of-a-kind model in Llama 3.1 and the brand spankin new 405B model.

What's it all mean? We gotchyu.

Join the conversation and ask Jordan questions on Meta here.

Also on the pod today:

• Highlights of Llama 3.1 and 405b ✨
• Functionalities and User Interface Updates 🧑‍💻️
• Meta's Focus and Its Implication on AI and Business 🤔

It’ll be worth your 57 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

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

New AI Tool Spotlight – Indigo lets you deploy your own prompt library, FinFloh is a one-stop solution for B2B credit decisioning and VideoGen generate professional videos in seconds.

New Tuned Models — A team of two announced a tuned version of Llama's 3.1 70B model called Reflection 70B. Its creators said a more formal research paper is incoming, but early results show Reflection's advanced tuning technique significantly enhances its performance, allowing it to outperform top competitors like GPT-4o on various benchmarks such as MMLU and GSM8K.

Trending in AI – An OpenAI investor said that Musk’s new AI chip cluster is ‘table stakes.’

Google – Google has added the ability to use fine-tuned Gemini models via an API key.

Money in AI - Salesforce has acquired data management firm Own for $1.9 billion.

AI in Medical – Google DeepMind has created AlphaProteo, an AI system that designs novel proteins.

AI in Media – The director of ‘Alien: Romulus’ is defending AI versions of dead actors after facing backlash.

AI in Education – This local college in Floria is rolling out its first ever Artificial Intelligence degree.

1. Salesforce Introduces Agentforce 🤖

Marc Benioff, CEO of Salesforce, has announced the launch of Agentforce, a platform designed to create autonomous AI agents that will transform customer service operations. This strategic pivot comes in response to a critical blog post from Andreessen Horowitz, which raised concerns about Salesforce's competitiveness in the evolving landscape of generative AI.

With early testing at companies like Wiley and Open Table, Benioff expresses confidence that Salesforce is well-prepared for this AI-driven transformation.

2. Xockets Takes on Nvidia and Microsoft 🧑‍⚖️

Xockets, a Texas-based startup, is taking legal action against Nvidia and Microsoft, alleging significant patent infringement related to its data processing unit technology. The lawsuit claims that Nvidia’s DPUs, including BlueField and ConnectX, rely on Xockets’ patented innovations first demonstrated at a conference in 2015.

Additionally, Xockets accuses Microsoft of benefiting from Nvidia's alleged infringement while participating in a buying cartel aimed at suppressing tech prices.

3. Salesforce Unleashes New AI Supermodels 🦸

Salesforce just dropped a bombshell with the launch of two game-changing AI models, xGen-Sales and xLAM, aimed at turbocharging sales automation! These models promise to take over complex sales tasks with lightning speed and precision, raising eyebrows about the future of human sales teams.

Senior VP Adam Evans is hyping it up, claiming these tools are not just another feature but a total game-changer for businesses looking to boost efficiency.

4. Video Game Voice Actors Secure Major AI Agreements 🎮

After a month-long strike, video game performers have successfully negotiated agreements with 80 games regarding crucial AI protections. The Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) has established interim agreements that enhance wages and provide safeguards against the unauthorized use of performers’ likenesses by AI.

While negotiations continue with major publishers like Disney and EA, independent studios such as Lightspeed L.A. are embracing these terms, allowing them to work with union talent

5. Global Tech Titans Unite for LLM Security 🤝

China’s top tech squad—Ant Group, Tencent, and Baidu—joined forces with US giants Microsoft, Google, and Meta to drop a groundbreaking standard for securing large language models (LLMs) at the Inclusion Conference in Shanghai. This “Large Language Model Security Requirements Supply Chain” aims to tackle risks like data leaks and model tampering throughout the entire lifecycle of these powerful AI tools.

With experts weighing in, this initiative is part of the World Digital Technology Academy’s push for comprehensive AI governance that’s more necessary than ever.

6. Honor Unveils AI Assistant to Compete with Tech Giants 🥊

Honor has made waves by introducing its enhanced AI assistant, Yoyo, designed to challenge industry leaders like Apple and Samsung. With the new Honor AI Agent, Yoyo can efficiently identify subscriptions on WeChat and Alipay, showcasing its advanced capabilities.

As competitors ramp up their digital assistants, Honor is positioning itself as a key player in the AI landscape.

Sora Competitor? Kling AI review and Runway comparisons

Are we witnessing the dawn of a new AI video champion?

Kling AI has finally been released globally! How well does it perform and can it compete with Runway?

We put it to the test. And you’re prolly gonna be surprised by the results.

🦾How You Can Leverage:


Meta is just the parent company of Facebook, right? 

Naaaah shorties. 

Meta and their new large language model updates could greatly shape the future of how we work. 

(Zero cap. This is for real.) 

About 24 hours ago, Meta flipped the AI world upside down with Llama 3.1 and us AI geeks are buzzing louder than a beehive on Red Bull.

An Open Source model that’s as good as the biggest frontier models in the world. 

What? 

It’s free. 

You can download it. 

Build off of it. 

So while there’s been plenty of talk since Meta stirred the AI pot, there’s also been a lotta oversight. 

Maybe too much focus on benchmarks, and not enough focus on what this does to the business world at large. 

So today, we cut through the chit-chat and gave it to you real. 

Here’s the tl;dr version of Llama 3.1: 

  • Meta released its ‘large’ LLM, the 405B parameter version 

  • Meta updated the small 8B and medium 70B versions that were released in April

  • There’s a lot of under-the-hood technical updates

  • These are all open-source (ish), and available today to use, download and build upon. (No waiting for multiple quarters for whatever Google teased. Lolz) 

Here’s an overview of what’s new, and the 5 things you need to know about Meta’s new models. 

Let’s Llama y’all. 👇

1 – Meta 405B and 3.1 Update = Impressive Specs 👀️

Yeah, yeah, yeah. 

We hear ya. Benchmarks?!?

Yeah. 

Even if you’re not a chart-reading, pencil-pushing benchmark lover like we are, we goooooooootttta talk about these benchmarks.

Not just because they’re crazy impressive. But because this is gonna set off a cheap compute race that everyone’s now gonna have to run. 

So check it: The 405B beast is scoring 88.6 on MMLU. 

For you non-nerds, that's like getting a 99.9% on the SAT of AI. It's breathing down GPT-4's neck (88.7). But here's the kicker y’all – it's open-source and Free 99!

And those "smaller" models? 

The 8B and 70B? 

They got jacked at the gym with 3.1. We're talking small-but-mighty, punching way above their weight class. The 8B is making Google's GEMMA 2B look like it skipped leg day.

Oh, and Meta's playing 4D chess with this "herd of experts" thing instead of the usual "mixture of experts."

It's like they took the AI playbook and said, "Nah, we're gonna write our own."

Try this:

Grab that 8B or 70B model on Hugging Face and run it on your laptop. Throw some mind-benders at it – like that apple and banana logic puzzle or the man and dog river-crossing thing.

See how it stacks up against the big boys.

Then get crazy – use it for something real in your biz and watch jaws drop. And you can run it locally. No internet. Private. And free. 

Whuuuuuuuuu. 

2 – Developer-Focused Strategy Against OpenAI 🎯

Alright, listen up devs.

Meta's not just entering the open source LLM game; they're flipping the whole dang table with the new Llama 3.1.

They've made Llama 3.1 available everywhere – AWS, Google Cloud, Microsoft, you name it.

It's like they're handing out superpowers at a dev conference.

(We don’t hate the Swag Table socks, but this is our kinda giveaway TBH.)

The timing? Chef's kiss. 

OpenAI drops GPT-4o Mini like 72 hours ago, thinking they're hot stuff. (I mean, that drop was dreamy.) 

Then BAM! Meta says, "Hold my beer startup” and releases this love llama beast. 

And OpenAI's response after seeing Meta release a free and open-source benchmarking beaut? They essentially said fine-tuning for free on us. (Seriously.) 

Try this: 

Let us know…. should we be focusing more on Meta?

Should we do a course on Llama?

Login or Subscribe to participate in polls.

3 – Llama 3.1 is Ready Today 🦙

No "coming soon" BS here, folks. 

Llama 3.1 is out in the wild, ready to party.

You can play with it on Meta.AI (just need a Facebook or Insta login – yeah, yeah, we know), or download Llama 3 and go ham on your own machine.

Sure, the interface ain't as pretty as ChatGPT.

But who needs pretty when you've got power? Switch between the 70B and 405B models like you're changing channels.

And hey, they even threw in an image generation feature for funsies.

Try this: 

Head to meta.ai or download whatever size your computer can handle.

Run some crazy prompts through it – like that business idea generator thing we mentioned.

See how it handles your half-baked, 2 AM creativity without you having to be a prompt engineer or even share data with Big Tech. 

(Personally, we hand out our data like candy at a parade, cuz we don’t care. Love us some great retargeting. But we get the whole ‘need for privacy’ thing.) 

Check out this tutorial when we gave Meta 3 the coding treatment. 👇

4 – An Open-Source Llama 3.1 is More Than Your Average LLM 😲

Llama 3.1 isn't just another chatbot on steroids.

It's like the Swiss Army knife of using AI to build AI. That’s super Meta, we know. 

(Double pun!)  

That model distillation feature that Zuckerberg talked about?

It's like having a mini-me maker for AI.

Take the big 405B model, tell it to make a smaller version that's a genius at just one thing your company needs.

Boom! Custom AI without needing a supercomputer.

(Alright, there’s some finetuning skills and dev brains needed in there, too, but you get the point.) 

And synthetic data generation?

It's like having a genie that grants unlimited wishes for training data. No more begging for datasets or scraping the internet like it's 2010.

Try this: 

Is Open source the way of the future?

Maybe.

Maybe not.

Meta’s Mark Zuckerberg thinks it is, and his long-form blog post on the future of Open Source AI is worth a read.

5 – Meta Llama Has Tremendous Business Impacts 💼

Buckle up, buttercup, 'cause this is where it gets real for your bottom line. 

Zuck's out here talking about a future with more AI agents than humans.

Which is exactly what we predicted in 2023 and people looked at us all funny. 

Every business running multiple AIs.

This ain't sci-fi; this is next week's memo from HR.

The AI gold rush just kicked off, and Meta's handing out free pickaxes, dynamite, and treasure maps.

We're not just talking about automating your boring spreadsheet tasks. We're talking AI that can handle the complex, "I need a human for this" kinda stuff.

The future of work?

It's not about being the smartest human in the room. It's about being the best human at working with AI.

It's prompt engineering meets business strategy, and it's coming for your job description.

Try this:

We think a lot about the future of work. And talk with industry experts almost every single day.

We’ve had some hot-takes on the future of work. Here’s an updated guide on the business side of AI right now.

⌚

Numbers to watch

September 9th

Apple’s iPhone 16 event will be September 9th, where many expect some updates on Apple’s AI integrations.

Now This …

Let us know your thoughts!

Vote to see live results

What Everyday AI source do you find the most valuable?

Login or Subscribe to participate in polls.

(Or your fave LLM like Claude, Gemini, Copilot, etc)

Reply

or to participate.