🌟 Your Personal Petaflop

Hey everybody,

I hope you had a great holiday period! This week’s newsletter will dive into Meta’s push for AI bots on its platforms, Nvidia’s bold AI announcements at CES, and OpenAI’s vision for superintelligence.

Quick Links

🤖 Meta Wants More AI Bots on Facebook and Instagram
Meta is bringing generative AI to its social-media platforms by introducing bots on Instagram and Facebook with profiles and content-creation abilities. These AI characters are designed to engage younger audiences and redefine interactions on the platforms. The move has raised questions about its practicality and whether it’s actually something that users want.

🫂 Software is mostly made of people
Thane Thomson wrote a good post on how software is much more than just code, it’s fundamentally about people. Running, value-producing software facilitates meaningful interactions between users and creators, improving lives in the process. Code is a vital layer but only one piece of the larger system that drives value and impact.

🧠 OpenAI is now working towards the ‘glorious future’ of superintelligence
OpenAI CEO Sam Altman revealed confidence in achieving artificial general intelligence (AGI), stating, “we know how to build AGI as we have traditionally understood it.” In a recent blog post, he predicted that AI agents could significantly impact company outputs this year. Altman also wrote about OpenAI’s focus on “superintelligence,” envisioning tools that could accelerate scientific discovery and innovation, driving unprecedented abundance and prosperity.

😂 Systems Ideas that Sound Good But Almost Never Work
Nine times out of ten, when someone says “let’s just,” what follows is far more complex than anticipated. Based on experience, these seemingly simple ideas rarely succeed because they underestimate the intricacies involved. The examples shared by Steven Sinofsky in this blog post are just a glimpse of countless situations where simplicity proves deceptive.

💯 Things we learned about LLMs in 2024
Simon Willison’s in-depth post, reflects on a transformative year for large language models. One standout observation was the ongoing ambiguity around “agents,” a term he finds frustratingly vague. Willison says “If you tell me that you are building ‘agents,’ you’ve conveyed almost no information,” which I agree with. There is a lack of a clear, shared definition despite its frequent use in the field.

Nvidia’s Personal AI Supercomputer

Nvidia has unveiled Project Digits, a personal AI supercomputer launching in May, powered by the cutting-edge GB10 Grace Blackwell Superchip. Compact enough to fit on a desk and priced from $3,000, this desktop-sized marvel can handle AI models with up to 200 billion parameters—processing power previously reserved for massive systems.

“AI will be mainstream in every application for every industry,” said Nvidia CEO Jensen Huang. “Project Digits empowers developers, researchers, and students to shape the age of AI.”

The GB10 chip delivers 1 petaflop of AI performance at FP4 precision, featuring Nvidia’s latest CUDA and Tensor Cores, linked to a 20-core Arm-based Grace CPU. MediaTek collaborated to enhance power efficiency.

Running on Linux-based Nvidia DGX OS, it supports frameworks like PyTorch and Python, while offering Nvidia's AI software suite, including NeMo and RAPIDS libraries. Developers can test locally and deploy seamlessly to cloud or data centers.

Nvidia was aiming to extend its momentum from 2024 into 2025 with a series of AI-focused announcements made at CES in Las Vegas yesterday. During the keynote, Huang also outlined a bold vision for AI-driven innovations, including software to empower robots and self-driving cars, and the introduction of Project Digits.

Here’s some clips from the talk too.

More Reading

• How I program with LLMs.

• Static search trees: 40x faster than binary search.

• The Joy of Under-Engineering.

• Create beautiful terminal-based code tutorials with syntax highlighting.

• The best way to iterate over a large array without blocking the main thread in Javascript.

Until next week,

Travis.