Nvidia's groundbreaking new chip & tools

Welcome to the 211 new subscribers who joined last week. I’m thrilled that you’re here to learn with me.

What’s brewing in AI #32

• Microsoft hires key people from Inflection AI
• Nvidia’s groundbreaking new AI chip and tools
• Why GPTs aren’t taking off (yet)
• The other AI news this week you don’t want to miss

‍

Dario’s Picks

I. Microsoft’s non-acquisiton of Inflection AI

‍

‍

Mustafa Suleyman, co-founder of Inflection AI (startup valued at 4$ billion), recently joined Microsoft as the head of the company’s s new AI divison, where he’ll be in charge of building and advancing Microsoft Copilot.

Suleyman isn’t the only one from Inflection Microsoft just hired; Chief Scientist Karén Simonyan and several other researchers from the startup join the newly started Microsoft division as well. The hiring was part of a 650$ million licensing deal in which Inflection’s models will be available on Azure; presumably as a way for the whole thing not to be seen as an acquisiton.

Inflection AI is the company behind the Pi (a more personal chatbot, which I’ve found myself using quite a bit). Pi will still be available but the company is changing it’s direction: Rather than focusing on Pi, the chatbot product, they’ll go B2B, focusing on API licensing and their AI studio, where commercial customers can test and fine tune their model to their own business.

Why it matters:

Inflection has actually created a great product with Pi (test it and see for yourself), including training their own LLM and acquiring the GPUs needed for that, but is now losing key talent. It makes sense that they’ll focus more on the big fish (businesses rather than consumers) to license their existing products, which would likely be an easier game than trying to compete with ChatGPT.

On the flipside, Microsoft is getting Inflection’s LLMs available on Azure, which could potentially be a big helper in making their copilots more useful, as well as new use cases for AI within Microsoft’s products that aren’t just about productivity.

‍

II. Nvidia’s new AI chip and tools

‍

‍

Nvidia announced new tools to power next-generation AI models:

• NVIDIA Blackwell: a new GPU with 2.5x performance of its predecessor. The GB200 version will be the world’s most powerful chip.
• NIM microservices, pre-built packages that enable plug-and-play creation custom AI tools.
• Omniverse Cloud APIs which expands the capabilities for simulation
• GR00T: a foundation model for humanoid robots.

Why it matters: Better, more capable GPUs expand the limits of what’s possible, and has the potential to drive the AI revolution forward much faster. The software products they’re launching in parallel to this tell a lot about where we’re headed next: easy-to-build AI agents, virtual worlds and humanoid robots.

‍

III. Character.ai gets voices

‍

‍

Character AI just introduced a new feature that allows users hear characters speaking. The move is part of the company’s plan to build a multimodal interface.

The chatbot platform now has a voice library (community-created voices that can be used in-chat) and a tool to create your own voice (by uploading an audio sample). When chatting to a character, users can now explore and switch between voices which are created by others and set to “public”.

Why it matters:

Character AI just took the voice-modality inside chatbots to the next level. I think we’ll see other popular chatbots take heavy inspiration from this soon. How about ChatGPT with your very own voice?

Want to try Character AI? Check out my categorised list of top c.ai bots.

‍

In Focus

Why GPTs aren’t taking off (yet)

OpenAI’s GPT store is off to a slow start, according to The Information.  Developers cite lack of analytics about their users, access being limited to paying customers (ChatGPT plus subscribers) and a promised monetisation program for creators that has yet to materialise.

‍

My take on this:

I agree with the factors mentioned above – I would add that a lot of advanced GPTs (the most popular ones are generally “advanced”) currently rely on sending users to external websites for the core functionality to work, which is a real friction in the user experience. Something more intuitive would be great.

My guess is that OpenAI is still tinkering with the whole concept of GPTs internally, and waiting with solving some of these limitations and weaknesses until their next big product update. If OpenAI’s concern was achieving many times more traction and interest in GPTs quickly, I think they would’ve already done so by making them available on ChatGPT’s free version.

GPT-5 is rumoured to not only be materially better than it’s predecessor, but also great at calling agents that can do work autonomously. To me, this sounds like it could be the next iteration of GPTs.

I have little clue why the monetisation program isn’t introduced yet (was planned for Q1). OpenAI is likely taking learnings from Poe (Quora’s own chatbot platform) which introduced a monetisation scheme 5 months ago that is similar to usage-based builder revenue program described by Sam Altman on DevDay back in November. Maybe the proposed revenue model isn’t providing enough value to developers, users or the company?

‍

GPTs

Top new arrivals in the GPT store

Highly rated, new GPTs featured in OpenAI’s official GPT store (from the last week)

1. MapGPT   Featured
2. Question Maker   Featured
3. Library of Babel  Featured

‍

Fresh off the whatplugin blog

I’m collaborating with writers to create easy tutorials and reviews of GPTs to achieve different tasks. Here’s what we have this week:

• Summarising PDFs with AI By Denise
• GPTs for Astrology By Raphael

We’re still working on improving the content and format of these articles. Any specific topic you’d like to see covered by us? Leave your feedback in the poll at the end of this email.

‍

Featured GPTs (sponsored)

Internship GPT

💬80+   ⭐️4.5

Internship-focused resume and cover letter creator

‍

‍

‍

❞ Graduating soon? This GPT helps you with your internship application. You tell it what job you’re applying for and it asks you specific questions about you and the role. Finally, it comes up with a tailored cover letter (free sign-up required to see it).

Submit your GPT to whatplugin and get featured in the newsletter.

‍

Bytes

• Emad Mostaque, Stability AI’s CEO and board member, resigns from his positions. Some of the company’s key researchers also left last week. Stability AI is well known for its open-source tools, particularly Stable Diffusion.
• Insightful comparison of ChatGPT, Claude and Gemini by Ethan Mollick.
• Worldcoin, a company backed by Sam Altman, open-sources part of Orb’s software. Orb is an eyeball scanner that lets people prove their humanness online.
• OpenAI is pitching Sora to Hollywood film studios. Meetings are reportedly ongoing this week.

Product

• OpenInterpreter launches 01 Light: a voice interface for your home computer.
• Stability drops new model that can turn single, flat images into 3D.
• Gemini Pro 1.5 (yup, the one with 1M context length) is available for everyone to try inside Google’s AI Studio. Coming soon to API, too.

‍

That’s a wrap for this week!

Fellow sorcerers – join me on LinkedIn.

Until next time,

Dario Chincha 🧙🏼‍♂️