🤖 This week’s AI news!

ALSO: Elon Musk’s powerful GPU farm + AI Conference

Estimated Read Time: 5 minutes

The future of AI is looking bright as major AI companies prepare for the big AI Conference in a few days. Plus we’re changing our release schedule to be more often and consistent starting Monday!

• ⚡️ Elon Musk activates the most powerful GPU farm.

• 🎥 The AI Conference is coming up.

• 💰 Former OpenAI cofounder raises $1B for AI safety.

• 🍓 OpenAI’s “Project Strawberry” is coming soon.

• 🗣️ Details on Amazon’s “new” Alexa.

⚡️ Elon Musk activates the most powerful GPU farm

What happened: Elon Musk just launched “Colossus”, the most powerful GPU farm in the world consisting of 100,000 Nvidia H100 GPUs. And he’s revealed plans to double the size of Colossus once more.

The details: 

• The project uses 100,000 H100 Nvidia GPUs connected into a single network. Elon Musk claims it’s the most powerful GPU farm in the world.

• xAI plans to expand Colossus to 200,000 GPUs, doubling its size. 50k of those GPUs will be the new Nvidia H200s, the rest will be more H100s.

• The estimated cost to build Colossus lands between $3B and $4B. And it only took around 120 days to build.

• Colossus will be used to train Grok 3. 😎

• The GPU farm is also extremely expensive to run as it costs at least 150MW of power. This is why nuclear power may become necessary for future AI data centers.

• Apparently Grok 2, which boasts serious performance and competes with top models, was only trained on 15,000 GPUs.

• It’s important to note that while AI training is expensive and very energy draining, pure power alone is not enough. Training data remains the next most important asset, one which AI companies still struggle to obtain legally.

ALSO:

• xAI recently released Groq LLaVA v1.5 7B, a multimodal AI model which understands both images and text. It supposedly runs 4x faster than GPT-4o.

• The new model answers questions about images and understands conversations involving text, voice, and pictures.

• The model is currently available as a free preview for developers.

Why it matters: This massive project shows the future direction for AI projects. I expect every major AI company to build similar massive GPU farms for training their models. And equally large data centers to run those models. The question is how will we produce enough energy for these data centers to run?

🎥 The AI Conference is coming up

• The AI conference will be on September 10th and 11th.

• The conference will feature multiple speakers from the biggest companies in AI, including OpenAI, Meta and Google.

• You can sign up for the conference here (it’s paid tickets though).

• We’ll be covering the conference’s events next week!

🧑‍⚕️ Google Deepmind’s AlphaProteo could be a breakthrough for cancer research. The model designs custom proteins to bind to specific targets… this could include cancer cells. Apparently it’s up to 300 times better at creating these proteins than current methods.

🦾 OpenAI is also developing new chips. These will be the company’s first in-house AI chips. Reportedly they will use TSMC’s advanced 1.6mm A16 process node and partner with other major companies, such as Apple, for chip design.

🥊 Nvidia might be undergoing an anti-trust subpoena. Reports have said Nvidia was served a subpoena by the Justice Department. Nvidia has since denied any such subpoena happening. Nvidia claims they simply spoke with the Justice Department.

🤖 1X’s new NEO robot is extremely human-like. The NEO Beta is the most human-like robot we’ve seen so far. It’s been designed for helping in household tasks and moves very quickly compared to other models. Check out a video here.

💰 Former OpenAI cofounder raises $1B for AI safety

What happened: Ilya Sutskever’s AI company, Safe Superintelligence (SSI), just raised $1B in funding.

The details:

• The goal of SSI is to build safe AI systems which surpass human intelligence.

• The company is willing to spend years on R&D and ensuring proper safety protocols before bringing a product to market.

• 🤑 SSI is already valuated at $5B despite starting 3 months ago.

• Ilya Sutskever is a former OpenAI cofounder. He was ousted not too long ago amid rumors that he and Sam Altman, who has become notorious for ignoring AI safety and dismissing his AI teams, were having an internal power struggle.

Why it matters: This is the first AI startup to receive so much money in such a short period of time for their seed funding. This is probably thanks to Sutskever, who helped make OpenAI what it is today. But AI is becoming more important everyday. And we need safe systems to ensure AI doesn’t become a tool for abuse, crime, or become a real life SkyNet. Unfortunately, OpenAI itself seems like they don’t care, so companies like SSI have to pick up the slack.

🍓 OpenAI’s “Project Strawberry” is coming soon

What happened: You remember Project Q*? The mysterious project which leakers said was a “threat to humanity” is now Project Strawberry; and it’s preparing to release soon according to recent reports.

The details: 

• Strawberry may join ChatGPT as soon as this fall and would be a significant leap over GPT-4o. It’s unknown in what capacity it will join ChatGPT.

• OpenAI Japan just teased details for GPT-Next. It’ll apparently be 100x faster than the current model without hogging more resources. It’s unknown if this will have anything to do with Strawberry though.

• You’ll be able to give Strawberry extra time to create an answer, which should give higher quality results. Apparently it’s quite good at tackling subjective content (I’ll believe it when I see it).

• Strawberry is being used to generate training data for another LLM project known as “Orion”. We don’t know much about Orion though. OpenAI’s Japan branch confirmed Orion will arrive next year.

• And apparently OpenAI is worth $100B. Yup. A recent valuation found the company is worth quite a bit as the company prepares for another funding round. Pay attention investors. That’s more than the GDP of some countries.

Why this matters: An AI that creates training data for other AI models is extremely valuable. This could be a game-changer and help future AI models complete multi-step tasks more easily. Not to mention we might get another high end model to use this year.

🗣️ Details on Amazon’s “new” Alexa

What happened: Amazon’s “Remarkable” Alexa update is coming next month and will reportedly use Anthropic’s Claude models.

The details:

• Reportedly the initial version of Alexa used Amazon’s own in-house AI models and was disappointingly slow and unresponsive. Thus the switch to Claude.

• The main features will be the improved complexity of conversations, an improved shopping experience and new home automation features.

• Amazon will charge $5 to $10 monthly for the new version of Alexa. So it better be good. The classic version will remain free.

• Alexa recently came under fire for its biased take on politics. The AI is willing to give reasons to vote for Kamala Harris for President, but not Donald Trump. This has supposedly been changed, but it’s proof of the problems AI assistants will have; nobody wants an AI that can’t/won’t do something for arbitrary or political reasons. You can check out a video here.

Takeaway: Alexa is in over 400 million devices worldwide. I know some people who use it regularly. If Amazon can create an affordable and high quality upgrade, people will buy it. Assuming Alexa gives unbiased information. 😆

🤖 Tech comes alive!

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 ;)