🤖 Musk’s supercomputer and Grok 2/3 announced

ALSO: The biggest open source AI model ever

Estimated Read Time: 4 minutes

What a week! An almost-Presidential assination. The Tech world went down all at once. 👀 And now AI has officially gotten its biggest and best open-source model ever and Musk builds a massive supercomputer in preparation for Grok 2 and 3.

• 🤖 Elon builds massive supercomputer + Grok 3 in December.

• 🛑 OpenAI safety concerns pile up.

• 🦙 The biggest and best open-source AI model has released.

🤖 Elon builds the massive supercomputer + Grok 3 in December

What happened: Elon Musk announced that xAI’s supercomputer in Memphis, one of the largest in the world, began training early Monday morning and that Grok 3.0 is planned to release this December.

The details: 

• The supercomputer began training at 4:20 a.m. Monday and uses 100,000 Nvidia H100 GPUs.

• Nvidia’s H100 GPUs are considered the fastest currently available. They are vital for training AI systems, which is why Nvidia is so profitable.

• ⚡️ The electric bill is astronomical. Each H100 chip has the power usage equivalent to one household.

• The power bill has sparked some local community concerns over whether this is a good thing.

• The facility will create around 300 jobs and is 785,000-square-feet big. xAI is currently hiring in Memphis.

• Musk also confirmed, in an interview with Jordan Peterson, that Grok 2 will release next month and Grok 3 is planned for December. They will likely use the new AI facility to train.

And Tesla plans to ween off Nvidia with their Dojo project. Basically Dojo is Tesla’s supercomputer and is currently reliant on Nvidia’s GPU tech. The problem is Nvidia’s chips are so good and in high demand that it’s difficult to acquire them. 🤑 And expensive.

Elon Musk has already diverted $500M of chips away from Dojo to the xAI facility. (Remember, Elon Musk owns xAI and Tesla, but they are separate companies.) And he’s said repeatedly that Dojo must find a way to compete with Nvidia so as to ween off reliance on their tech. They have to.

"We do see a path to being competitive with Nvidia with Dojo. We kind of have no choice." - Elon Musk

💰 Musk previously promised Tesla shareholders that Dojo could add $500B to their market cap. It’s a big deal.

🎮 AI helped create the NCAA football game. Videogame company EA used AI to create the 3D avatars of 11,000 real college football athletes. The game employs AI technology to create digital avatars from photos. Simple and effective. It’s been 11 years since their last NCAA game. EA mentioned that any imperfections or problems resulting from the AI generation were easily fixed by manually tweaking the models. Take notes, guys. This is how you use AI to make lots of stuff efficiently.

🛑 OpenAI safety concerns pile up

What happened: A group of five U.S. senators has demanded that OpenAI provide data on how it plans to meet its safety and security commitments. All of this as a result of the recent whistleblower incident.

The details:

• Five senators sent a letter to OpenAI CEO Sam Altman. The group includes four Democrats and one Independent lawmaker.

• The concern is pretty basic; the senators want actual (data) proof that OpenAI will not misuse their AI.

• OpenAI was given a deadline of August 13 to respond.

• Whistleblowers recently filed complaints saying they were forced into unfair and dangerous NDAs which prevented safe, legal whistleblowing. This included docking their pay if they didn’t get the company’s approval before releasing information.

• OpenAI also reassigned AI Safety Team leader Madry. Joaquin Quinonero Candela and Lilian Weng will lead the preparedness team temporarily. It’s unknown why or what he’s working on.

• The preparedness team assesses the readiness of OpenAI’s AI models for artificial general intelligence.

Why it matters: Another day, another OpenAI controversy. It’s a shame the face of AI is always such a mess. But insider reports, leaks, and former employees all point to Sam Altman being obsessed with progress at the expense of safety. OpenAI makes a great case for government regulation.

🦙 The biggest and best open-source AI model has released

What happened: Llama 3.1 405B just released! It’s the biggest and best open-source AI model we’ve seen. And Meta (the creators of Llama) claim it surpasses top competitors like GPT-4o and Claude 3.5 Sonnet.

The details: 

• Llama 3.1 has 405 billion parameters, surpassing any other open-source models. There’re also 70B and 8B variants.

• The new model is over 5 times larger than the previously released versions.

• Meta believes the accessibility and power of their new model will make Llama 3.1 the most popular AI by year’s end.

• Training the AI involved over 16,000 Nvidia H100 GPUs.

• A new feature allows image generation based on specific likenesses.

• Here are Llama’s license agreements. Check them out if you’re interested in using this for development.

• The 405B model is free but limited with high volume usage.

And of course, Meta is tight-lipped about the data they used for training. I’d wager there’s some copyright infringement waiting to be discovered.

Why this matters: This is big. Open-source is now the frontier for AI development. This is gonna shakeup the AI industry in a good way.

This model so large and powerful that every third party app we use will be powered by Llama. The only problem is the volume limit on the 405B variant. Hopefully they resolve that soon.

🦙 More llamas?

Source: Midjourney

Give us your feedback!

Got feedback? Rate today’s newsletter by clicking below! 

Got more personal feedback? Fill in the box when you rate us and give us your criticism and feedback.

Thank you for reading! 

❤️Share the (ARTIFICIAL) Love!

Got a friend who wants to appear as smart as you? An AI fanatic or someone shaking in their boots of getting replaced? Share us by clicking the link below and you’ll get our love ;)