Ilya's Company Raises $1B in 3 Months

Another OpenAI Founder looks to go big with his own venture and Google Gemini’s Custom Gems is overtaking Custom GPTs…

Building a Safer AI Future - Ilya's SSI Gets $1B

OpenAI co-founder Ilya Sutskever has raised $1 billion for his new startup, Safe Superintelligence Inc. (SSI), which is dedicated to developing safer, more controlled AI systems. This venture highlights the increasing focus on AI safety amid growing concerns about the unchecked development of AI technologies.

• The innovative startup is fully dedicated to creating safer and more controlled AI systems in a "straight shot, with one focus, one goal, and one product.”

• Several high-profile venture capital firms and tech giants such as Andreessen Horowitz, Sequoia Capital, DST Global, SV Angel, NFDG, and others are backing SSI. Investors are clearly placing their bets on a future where the safe deployment of AI becomes just as important, if not more, than sheer advancement.

So What?

As you may recall, Ilya was influential in Sam Altman’s ouster from OpenAI in November 2023, and once he departed the company himself in May, the superintelligence operation severely tanked. With heightened global scrutiny on AI, SSI could reshape the industry’s priorities toward responsible innovation.

Let’s Build Personal Assistants with Gemini’s Custom Gems

Last week, Google's Gemini received a significant upgrade with "Custom Gems," a feature that allows users to create personalized AI chatbots tailored to specific needs, akin to ChatGPT's Custom GPTs. Here’s how you can create your own:

1. Log onto the Gemini homepage. Currently, Custom Gems can only accessed by Gemini Advanced users (you can use the free trial).

2. Navigate to “Gem Manager” on the bottom left, it is an icon that looks like a jewel.

3. The Gem Manager has premade assistants made by Google you can play around with, but to make your own, click on “New Gem.”

4. Name your gem, give it instructions. You can use the “@” symbol to give your assistant access to certain Google products like Docs, Gmail, and Drive.

5. Give it a spin!

Pro Tip 🔥: On the bottom of the Instructions where you create your Custom Gem, there is a pencil icon you can use to create an AI-generated set of instructions.

Microsoft Says “Don’t Forget About Us” And Reminds us Phi-3-Vision is Still Awesome

Microsoft (re)introduced Phi-3-Vision at a recent keynote event from their research group. Jianfeng Gao emphasized how it’s an advanced AND economical open-source multimodal model.

• As a member of the Phi-3 model family, Phi-3-Vision enhances language models by integrating multi-sensory skills, seamlessly combining language and vision capabilities.

• The model is designed to be accessible and affordable, and it’s available on platforms such as Hugging Face.

So What?

Building with small language models (SLMs) are highly cost effective. Now that we have multimodal capabilities with something like Phi-3-Vision, it unlocks more possibilities for AI practitioners. Microsoft also mentioned that this specific model can be a stepping stone to a “language-vision-action” model, which will also be able to take action on a human’s behalf.

How Might LLMs Store Facts?

This excellent video from 3Blue1Brown on YouTube talks about how large language models (LLMs) store facts.

Remember, large language models (LLMs) are a type of AI that are trained on a massive amount of text data. This allows them to generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.

He uses the example of a large language model being able to predict that the sport Michael Jordan plays is basketball. The model does this by having a high dimensional space where different directions encode different types of meaning.  

My breakdown:

• In simpler terms, the model uses internal "knobs and dials" to represent words and concepts.

• The activation patterns of these knobs allow the model to make connections and predictions.

Stay updated with our favorite highlights, dive in for a full flavor of the coverage!

OpenAI’s lead Japan executive, Tadao Nagasaki, teased the upcoming “GPT-Next” model. The new model is expected to further enhance the performance of OpenAI’s model capabilities by 100x.

The UK is set to open its first “teacherless” AI classroom at David Game College in London. This innovative classroom will rely entirely on artificial intelligence to teach students, offering a glimpse into the future of learning environments.

An AI model is providing NFL betting picks for Week 1 of the 2024 season, including predictions against the spread, over/under, and money line outcomes.

There’s an AI Agent Minecraft server out there… The first ever agent civilization boasts 1000+ truly autonomous agents creating their own culture, economy, religion, and government.

Anthropic announced the Claude Enterprise plan to help organizations securely collaborate with Claude using internal knowledge. The plan offers an expanded 500K context window, more usage capacity, and a native GitHub integration to work on entire codebases with Claude.

The perfect place to come engage with us at Digestible AI!

If you have any suggestions or specific feedback, simply reply to this email or fill out this form. Additionally, if you found this insightful, don't hesitate to engage with us on our socials and forward this over to your friends!

You can find our past newsletter editions here!

This newsletter was brought to you by Digestible AI. Was this forwarded to you? Subscribe here to get your AI fix delivered right to your inbox. 👋