🤯 OpenAI’s Strawberry is here!

ALSO: The godmother of AI raises $200M

Estimated Read Time: 4 minutes

OpenAI has released their latest AI model… and it’s a doozy. We’re talking about a massive leap in IQ of 30 points and massive improvements to general reasoning capability. We go over EVERYTHING with OpenAI’s new model, and it’s the focus of today’s newsletter. Let’s dive in!

• 🤯 OpenAI’s GPT-o1 (Strawberry) is powerful!!!

• 🫠 OpenAI may become for-profit next year.

• 💰 The “Godmother” of AI raised $200M for a new startup.

🦾 OpenAI’s GPT-o1 is a powerful leap forward

What happened: OpenAI has released a new AI model called o1, the first in a series of “reasoning” models designed to solve AI’s biggest problem… having common sense. Alongside o1, a smaller and more affordable version named o1-mini has also been introduced. Oh, and GPT-4 has a mini version available now too.

The details: 

• o1 is trained using a completely new algorithm and dataset aimed at creating better “chain of thought” reasoning. Again, trying to give AI common sense and reasoning.

• The “chain of thought” process tries to mimic human-like step-by-step reasoning. So it should understand context better, allowing the model to tackle more complicated tasks. It should be better at completing simple tasks too.

• o1 hallucinates less and is more accurate in reasoning tasks.

• But GPT-o1 doesn’t perform as well on factual knowledge and lacks features like web browsing and image processing. But fewer hallucinations are a big deal.

• o1 has typical improvements as well, such as greater mathematical and coding ability.

• Unlike earlier models, o1 shows its reasoning steps, such as “thinking”.

• GPT-o1 is still considered a “preview”, so it’s in early stages of development.

• The name change is supposed to represent bringing AI back to the ground floor so it can build up better than before.

• OpenAI claims the next version of o1 will perform at the same level as PHD students on challenging tasks like physics, chemistry, and biology.

• Consumer Pricing: ChatGPT Plus and Team users get immediate access, with plans to extend o1-mini to all free ChatGPT users.

• Dev Costs: Developer access is more expensive than previous models, costing $15 per 1 million input tokens and $60 per 1 million output tokens.

• This doesn’t replace GPT-4. They will coexist and GPT-4 is still available.

• Yes, this is indeed the Strawberry model the rumors talked about. Wasn’t long ago we reported the rumors that a new model would release soon. An even more powerful model should release next year.

The other models:

• o1-mini is a scaled-down version of the o1 model, meaning it has fewer parameters and is less computationally intensive. This makes it more accessible for users who don’t require the full power of the standard o1 model. But it’s also cheaper for developers to use. I expect this’ll be popular for developers.

• Mini is sort of faster. o1-mini processes queries faster than the full o1 model. Its reduced complexity allows for quicker response times, making it suitable for applications where speed is crucial.

• o1-mini will eventually be available to all ChatGPT users, free and paid.

• The model generally won’t perform as well on complex tasks as the full o1 model. Which… is to be expected honestly.

• Basically all the same stuff that applied to GPT-4o mini.

Why it matters: That’s a lot. Human-like reasoning has been the biggest weakness of all AI models. They’re smart… except they don’t understand anything. GPT-o1 is new and everyone’s still breaking it in, but early results are clear; the model is WAY smarter than before and it’s easier to get tasks you want done. This is a massive step forward for AI. Possibly even historic, actually.

💰 OpenAI may become fully for-profit. A large part of the deal to get their recent funding was removing the profit cap for investors. According to Fortune, Sam Altman recently told employees at OpenAI the company will transition to be closer to a traditional-for profit business. So I guess all of those concerns about the direction of the company were true? 🫠

🇬🇧 Meta is resuming efforts to train its AI systems using public Facebook and Instagram posts from U.K. users after incorporating regulatory feedback into a revised opt-out approach. Starting next week, U.K. users will receive in-app notifications explaining the changes, and Meta plans to use public content for AI training in the coming months. So basically the same situation as the USA and Australia. Well, not really. UK users will have some sort of simplified opt-out process, but it’s unclear what that means. Meanwhile EU users are safe… for now.

💰 The “Godmother” of AI raised $200M for a new startup

What happened: Fei-Fei Li, renowned as the “godmother of AI,” has raised $230 million for her startup World Labs, aiming to develop AI technology that comprehends the three-dimensional physical world. Pretty important in an era full of robotics companies trying to enter the workforce.

The details:

• “Spacial intelligence” is the company’s focus. Basically the AI’s ability to reason about 3D environments.

• World Labs received $230 million in initial funding led by Andreessen Horowitz, New Enterprise Associates, and Radical Ventures, with additional investments from AMD Ventures, Intel Capital, and Nvidia’s NVentures. Everyone’s in on this.

• The startup aims to create AI models known as “large world models” (LWMs) that understand the structure of the 3D world, addressing limitations in current generative AI models.

• Who is Li? Li developed ImageNet, a large-scale image dataset that has been foundational in the advancement of computer vision. She’s also a Professor at Stanford and has an influential hand in educating about AI.

• Alongside Fei-Fei Li, the founders include computer vision researchers Justin Johnson, Christoph Lassner, and Ben Mildenhall, all experts in the field.

• Li will maintain her role at Stanford University’s Human-Centered AI Institute while leading World Labs.

Why it matters: This is clearly aimed at VR and robotics companies. One of their biggest issues is figuring out how to navigate the real world. Robotics companies are creating their own solutions… but whoever solves this problem and sells it to everyone stands to make a lot of money. It’s like saying the person who creates gasoline will profit from the car industry.

🐴 Horses!

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 ;)