šŸ˜ŗ Next ChatGPT is Coming

PLUS: OpenAI's ex-board members SPEAK OUT!

Welcome, humans.

The wildest part of Nvidia's meteoric stock surge isn't just its 30x increase over the past 5 years, but the fact that weā€™re convinced almost no $NVDA investors can actually explain what the company does.

Seriously, we spoke to someone recently who thought Nvidia chips were the latest Pringles flavor. Okay, thatā€™s not entirely true, but wouldn't it be hilarious if Robinhood created a feature requiring investors to pass a Nvidia quiz before buying the stock?

And hey Pringles, if you do whip up a Nvidia-flavored potato chip, be sure to give credit where credit is due (in the form of royalties, obviously)!

Hereā€™s what you need to know about AI today:

  • OpenAI is training the next version of ChatGPT.

  • This new model is built on mega-powerful Microsoft hardware.

  • 7% of Americans use ChatGPT every day.

  • Gemini 1.5 Pro + Advanced passed ChatGPT-4-Turbo on the LMSYS leaderboard.

On last weekā€™s podcast: Pete explains Anthropicā€™s big research breakthrough that maps the mind of AI models (Apple Podcasts, Spotify, YouTube).

OpenAI is prepping the next version of ChatGPT.

A year ago, amid the buzz of ChatGPT-4's release, we began to worry that AI might be advancing too fast for the rest of the world to keep up.

The thought was: if ChatGPT-4 launched shortly after ChatGPT-3.5, then we could see a version far superior to GPT-4 emerge just as quickly.

So we waited. And waited. Yet no ChatGPT-5.

Instead, a year later, we got ChatGPT-4o, a significant upgrade over GPT-4, though not as drastic as the leap from GPT-3.5 to GPT-4.

created by ChatGPT-4o

Yesterday, OpenAI confirmed that itā€™s actively training the next version of ChatGPT. 

The modelā€™s name is TBDā€”OpenAI hinted that it might not be GPT-5ā€”but itā€™s teased by Sam Altman and others to blow all other models out of the water in multiple dimensions.

We also know that itā€™s being trained on Microsoft hardware orders of magnitude more powerful than previous hardware:

Why it matters: No one knows the exact arrival of this model, but we're eyeing early 2025, considering:

  1. OpenAI likes to take its sweet time between model releases.

  2. Training and fine-tuning new models typically take at least several months each.

  3. GPT-4o launched about a year after GPT-4, following a similar gap after GPT-3.5.

This suggests that ChatGPT-4o, with its full voice and vision capabilities, will likely be the best AI available through the end of 2024.

Also coming with GPT-5: New safety efforts, according to OpenAI.

Itā€™s forming a new Safety and Security Committee featuring Board members Bret Taylor, Adam Dā€™Angelo, Nicole Seligman, and Sam Altman (shocker) to ensure that the AI itā€™s building doesnā€™t kill us all.

Of course, this all comes less than two weeks after OpenAI dissolved its long-term AI risks team after both of the teamā€™s leaders resigned. Jan Leike, one of the leaders who resigned and has since criticized OpenAIā€™s commitment to AI safety, just joined Anthropic.

Check out Peteā€™s analysis of the safety debate happening inside of OpenAI (Apple Podcasts, Spotify, & YouTube)!

FROM OUR PARTNERS

Hire a world-class AI team trusted by Fortune 500 companies and leading startups. 

Building AI software in your company is hard. Youā€™ve got three options:

  1. Track down technical engineers and pay them a fortune.

  2. Try to DIY, but building AI products is highly technical. 

  3. Use AE Studio

Their stacked team of AI experts builds the perfect AI solution for your business. Thatā€™s why theyā€™re the go-to for leading startups and Fortune 500 companies.

Their development, design, and data science teams work closely with your team to make sure your AI does exactly what you need it to. 

Around the Horn.

  • Gemini 1.5 Pro and Advanced now rank as the 2nd and 3rd most powerful models on the LMSYS chatbot leaderboard behind GPT-4o.

  • Elon Musk wants to build a 10,000-semiconductor supercomputer to train xAIā€™s upcoming Grok model.

  • ChatGPT has yet to decrease the # of visitors Wikipedia receives.

  • 47% of Americans have heard of ChatGPT, with 7% using it daily.

Treats To Try.

  1. Coactive AI uses multimodal AI to help businesses find and analyze visual content more easily (raised $30M).

  2. Kabata is an AI-powered dumbbell that counts your reps, corrects your form, and recommends weight changes (raised $5M).

  3. QA.tech is an AI that tests your web apps, providing bug reports and actionable feedback (raised $2.5M).

  4. Skej is an AI email assistant that schedules meetings with your contacts based on your calendar and preferences.

A Cat's Commentary.

Where do you #Neuron?

Send us a picture of where you read The Neuron here (with some spicy commentary), and you might get featured next week!

Lakshmi from Trinidad and Tobago: ā€œIt's almost Friday, may I curl up into a ball too please?ā€

Alex from Arnold, California: ā€œKeeping up with all the AI news remotely at our cabin in the woods!ā€

Gerardo from Garland, Texas: ā€œCatching up on Neuron during my daughterā€™s gymnastics hour.ā€

From Green Bay, Wisconsin: ā€œHome of the 13x World Champion Green Bay Packers.ā€

Thatā€™s all for today, for more AI treats, check out our website.

Get your brand in front of 450,000+ professionals here.

See you cool cats on Twitter: @nonmayorpete & @noahedelman02

What'd you think of today's email?

Login or Subscribe to participate in polls.