😺NVIDIA's AI Superbowl

PLUS: Prompt tips to make AI writing more natural

Welcome, humans.

OpenAI Chief Product Officer Kevin Weill just declared that “This is the year that AI gets better than humans at programming, forever. And there's no going back.”

Bold claim, Kevin! Meanwhile, Reddit's software engineers collectively rolled their eyes so hard they probably saw their own brains. One user perfectly summed it up: “Product Officer says Product Officer stuff.”

That said, there’s a reason all the major AI labs like OpenAI, Anthropic, Google DeepMind, and even upstarts like Poolside are focusing on making AI better than humans at making software—because, as Kevin says, “with software you can create anything you want, almost.”

Take this guy who built a local AI connected to a tax regulation database that finally convinced his wife his “silly computer hobby” was useful. “A small step for AI, a giant leap for household credibility,” he wrote, while tax-burdened Redditors offered up limbs to get access to his setup.

Also: If AI really does replace human programmers this year, shouldn't OpenAI be the first to put their money where their mouth is and let AI write 100% of their code? Although, are we 100% sure they aren’t already doing that?

Here’s what you need to know about AI today:

NVIDIA made a slew of announcements at its “Super Bowl of AI.”
Claude prototyped voice mode with potential ElevenLabs partnership.
Meta's Llama hit 1B downloads, up 53% since December.
Spain released first all-AI album; Italy published first all-AI newspaper.

Advertise in The Neuron here.

NVIDIA’s GTC event featured AI Factories, Robot Workers, and a Trillion-Dollar Data Center Future

As NVIDIA CEO Jensen Huang took the stage at GTC 2025 yesterday (in his iconic leather jacket, of course), he kicked off his keynote with a bold declaration: “This is the Super Bowl of AI,” joking that “everybody wins at this Super Bowl.”

Well, except maybe NVDA stock, which dipped slightly during his presentation (here’s why). To be fair, Jensen did call himself “chief revenue destroyer” mid-keynote…

That’s despite a flood of new announcements (here’s a 16 min video recap), which included:

A new architecture for massive AI data centers (now called “AI factories”).
A physics engine for robot training built with Disney and DeepMind.
A partnership with GM to develop next-gen vehicles, factories and robots.
A new Blackwell chip with “Dynamo” software that makes AI reasoning 40x faster than previous generations.
A new “Rubin” chip slated for 2026 and a “Feynman” chip set for 2028.

For enterprises, NVIDIA unveiled DGX Spark and DGX Station—Jensen's vision of AI-era computing, bringing NVIDIA’s powerful Blackwell chip directly to your desk.

The DGX Spark, about the size of a small book, will let AI developers run sophisticated AI models locally that previously required massive data centers (that is, run up to 200B parameter models locally).

On robotics, which Jensen claimed “could very well likely be the largest industry of all,” NVIDIA announced a comprehensive training ecosystem:

Omniverse provides the simulation environment.
Cosmos generates photorealistic training data (HF, Github).
Isaac Lab enables skill learning through imitation or reinforcement.
Newton offers realistic physics training.

The highlight was Isaac Gr00T N1, an open foundation model for humanoid robots (paper, model weights, code, data) that thinks “fast and slow” like humans do to perceive, plan, and act (video).

But the best reveal of all was probably this little guy:

That’s Disney’s BDX robot, which was trained with Newton. Disney will actually be testing these robots in several them park locations, too.

Jensen also predicted 100% of NVIDIA’s software engineers will be AI-assisted by year-end, and that robots will fill 50 million worker shortages by 2030—and could earn $50K annually to do it.

Yeah, uh we have some follow up questions—”worker shortage?” You mean from all the jobs that’ll get cut by AI??

Why it matters: NVIDIA is building the entire AI technology stack from chips to operating systems. Companies will likely run NVIDIA-powered AI models on NVIDIA hardware and train NVIDIA-powered robots using NVIDIA simulation tools within a decade—at least, according to NVIDIA.

The robotics vision is either exciting or terrifying, depending on your perspective. Will robot workers actually replace 50M human jobs by 2030? That timeline seems aggressive, but NVIDIA is clearly betting billions that the robot revolution is finally, actually here. And TBH, it does kinda seem like it’s about to go down…

FROM OUR PARTNERS

Here’s why Attention is all you need…

A little well-kept secret for companies crushing it with AI is that only a handful of AI tools are actually worth using.

That’s why we—Noah & Grant—use ChatGPT and Claude, along with a killer product called Attention.

Here’s how it works:

We tell Attention exactly what to pay attention to in our meeting (goals, budget, etc).
Attention listens in to our call.
After, Attention outputs important insights, action items, and follow-up emails.

There’s plenty more features, too. So if you sell anything, literally drop what you’re doing and go book a demo with Attention ASAP.

Prompt Tip of the Day

Check out this thread with a bunch of prompt advice for how to get ChatGPT to write more naturally. The simplest version is below.

1️⃣ Keep short and sharp.

2️⃣ Be concise and direct.

3️⃣ Make this mobile friendly to read.

4️⃣ Use language like I am talking.

Source

Treats To Try.

*GRID spreadsheet API is a spreadsheet engine optimized for AI and automation. Reliable, verifiable calculations—without hallucinations. Get started with GRID’s API.
Gemini Canvas is Gemini’s version of ChatGPT’s Canvas / Claude Artifacts tool that gives you an interactive workspace to create documents and code.
Aha automates your entire influencer marketing process with AI agents that find, negotiate with, and optimize campaigns for relevant creators.
Twos PALs transforms your written notes into actionable items with automatic reminders, task detection, and smart links for products, movies, and more (video).
Kintsugi tracks where you owe sales tax, calculates correct amounts, and files all your returns with one click.
Livebench is still probably the best benchmarking tool for testing whether or not a new AI model lives up to the hype—right now, 3.7 Sonnet thinking is in the lead, followed by ChatGPT o3-mini high and grok-3-thinking, and QwQ-32B is also considered just a bit better than DeepSeek-R1.
HuggingFace Daily Papers helps you search through AI research papers or find random papers with one click.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

This is ReCam Master, which you can try here, but it’ll take some time to get the video back—TrajectoryCrafter is the totally open source version of this (for non-commercial use), btw.

Claude is working on a voice mode with potential partners including ElevenLabs, and already has some prototypes.
Meta hit 1B downloads of its Llama models—up 53% since 650M downloads last December.
A US Federal appeals court agreed with the US Copyright Office that content created by AI alone can’t be copyrighted.
Spain sets record for first entirely AI music album and Italy sets the record for first newspaper edition produced entirely by AI.
The Arizona Supreme Court created AI Avatars to deliver rulings and connect with the public—the first such use by a state court system.

FROM OUR PARTNERS

Speechmatics

Transform your applications with speech-to-text that outperforms competitors by a staggering 25% in real-time accuracy.

Get 90+% precision in under one second across 55+ languages. Power your transcription, AI assistants, and medical documentation solutions without compromise.

Try Speechmatics free today.