Everyday AI
Posts
The key to autonomous driving: what's next

The key to autonomous driving: what's next

Federal copyright ruling on AI art, GPT-4.5 report on hallucinations, Italian newspaper publishes all AI issue.

Everyday AI
March 20, 2025

👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI

Outsmart The Future

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: Our guest today built robots for Mars and shared the secret unlock to autonomous vehicles. Go listen.

🕵️‍♂️ Fresh Finds: ChatGPT bug used in thousands of exploits, big NotebookLM updates, Apple facing lawsuits over AI delays and more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: OpenAI drops impressive audio model, Claude FINALLY gets web access, Perplexity seeks $1 billion funding. Read on for Byte Sized News.

🧠 Leverage AI: NVIDIA dropped a ton of update around autonomous vehicles. Go see what it means. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about Humanoids from an expert, Federal copyright ruling on AI art, GPT-4.5 report on hallucinations, Italian newspaper publishes all AI issue and more. Check it here!

Autonomous Driving: How new NVIDIA tech will make it a reality

You ever autonomously land a robot on Mars? 🪐

Marco Pavone has helped do just that. Marco is NVIDIA's Director, Autonomous Vehicle Research.

His next challenge? Bring true autonomy to vehicles on the road. 🚘

Yeah, yeah, yeah. We’ve been hearing that autonomous vehicles are coming for like a decade.

But new announcements at NVIDIA GTC are making that a reality.

Like…. This year.

Also on the pod today:

How Cosmos generates new worlds 🌏
Dashcam videos for the training win 📹
AI simulation breakthroughs 🦾

It’ll be worth your 33 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Spotify | Apple Podcasts |
Google Podcasts | Amazon Music |

Upcoming Everyday AI shows

How AI is making us more productive and dumber at the same time

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Epiphany turns your voice notes into actions, RecordAI is a WhatsApp friend that uses AI to remind you about tasks, Notebooks turns your content into a central hub for future content production.

AI and Weather — AI-powered weather forecasting is set to revolutionize predictions—faster, cheaper, and more accurate than ever. Curious how it works?

NotebookLM — NotebookLM just shipped a super useful feature to its platform in mindmaps.

🛳️Rolling out interactive Mindmaps in NotebookLM!
I'm so inspired by the Exploratorium here in SF - What if every notebook generated your own personal set of interactive understanding toys that help you learn through play? What if instead of text or images, Notebook could even
— Simon (@tokumin)
6:51 AM • Mar 19, 2025

AI in Telco – Nvidia is teaming up with big names like T-Mobile and Cisco to push AI-native 6G tech.

AI in legal — LexisNexis debuted Protégé, an AI model they developed based on Mistral.

AI and security — ChatGPT bug exploited in thousands of attacks—banks and sensitive data at risk.

AI Developers — xAI and Vercel have teamed up for some developer goodness:

Vercel and @xai are partnering to bring zero-friction AI to developers.
• Grok models are now available on Vercel
• Exclusive xAI free tier—no additional signup required
• Pay for what you use through your Vercel account
• Zero-config auth, API keys, and env vars
— Vercel (@vercel)
5:16 PM • Mar 20, 2025

Apple Intelligence — Apple faces a class-action lawsuit over delays in its promised AI features

1. OpenAI Unveils Advanced Voice Models for Developers 🎙️

OpenAI has launched three new cutting-edge voice AI models—gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-mini-tts—available now via API for developers who want to create apps with lifelike voice interactions.

These models promise better transcription accuracy, customizable emotional tones in speech, and real-time streaming capabilities, making them ideal for industries like customer service and meeting transcription. While the announcement is a major step forward, competition remains fierce, with rivals like ElevenLabs and Hume AI offering similar tools at competitive prices. OpenAI also teased a public contest on its demo site, OpenAI.fm, to showcase creative uses of its voice tech—complete with rare prizes for top entries.

2. Claude Finally Gets Web Smarts, years later �*

Anthropic’s AI chatbot, Claude, finally received web search capabilities, a major upgrade unveiled in preview for U.S. subscribers.

The feature, powered by the latest Claude 3.7 Sonnet model, allows users to toggle on web search and receive real-time, citation-backed responses from sources like NPR and Reuters. While this puts Claude in line with AI heavyweights like ChatGPT and Google’s Gemini, concerns linger over potential misinformation—an issue common across chatbots, according to studies by the Tow Center for Digital Journalism

3. Apple’s AI and Siri Stumbles Again Amid Executive Shakeup �*

Apple is delaying its much-hyped personalized AI features for Siri, a move that critics say highlights deeper struggles at the tech giant, according to a Bloomberg report.

CEO Tim Cook has reportedly shifted Siri development to Mike Rockwell, creator of the Vision Pro headset, signaling dissatisfaction with AI chief John Giannandrea’s progress. Analysts like Forrester’s Dipanjan Chatterjee argue that Siri has fallen behind rivals like Amazon Alexa and Google Assistant, especially with the rise of generative AI chatbots setting a new standard. While Apple searches for its next big innovation, questions loom about the company’s ability to keep pace in a rapidly evolving AI landscape.

(Our take: Apple’s Vision Pro was a disastrous rollout and launch. This is a weird move.)

4. Perplexity AI Eyes $18 Billion Valuation Amid Ambitious Expansion Plans 🤑

Perplexity AI is reportedly in early talks to raise up to $1 billion in funding, potentially valuing the AI search engine startup at $18 billion, according to Bloomberg.

Founded in 2022, the company has rapidly scaled, tripling its valuation twice in 2024 and achieving nearly $100 million in annual recurring revenue. Perplexity’s innovations include its AI-powered search tool, an internal file search product, and plans for a web browser and AI Phone collaboration with Deutsche Telekom.

5. Energy Meets AI: Major Players Form Open Power AI Consortium 🔌

A groundbreaking collaboration was unveiled at Nvidia’s developer conference, where energy giants and tech leaders like Microsoft, AWS, Oracle, and Nvidia joined forces to launch the Open Power AI Consortium, spearheaded by the Electric Power Research Institute (EPRI).

According to Fast Company, the consortium aims to leverage AI to enhance electric grid reliability, streamline energy management, and optimize power asset performance. While notable AI developers like Google and OpenAI are absent, over two dozen U.S. utility companies—including Con Edison and Pacific Gas & Electric—have signed on.

🦾How You Can Leverage:

Marco Pavone built robots for MARS.

Now he's telling us the wild truth about Earth's self-driving future at NVIDIA. Spoiler: it’ll happen sooner than you might think.

Marco is a Stanford professor and NVIDIA's lead autonomous vehicle researcher. He said autonomous driving is about more than collating a bajillion data points and running even more simulations.

Example: in Northern Italy a Blinking Headlight Means "Go Ahead" But 100 Miles South It Means "I'll Crash Into You.”

Turns out, understanding how Italians flash their headlights could matter more than sensor arrays. WILD.

The good news?

NVIDIA’s recent announcements at their GTC conference might mean autonomous driving is in reach.

(For real this time.)

Yeah, we’ve been hearing autonomous vehicles are around the corner for more than a decade.

But now, it’s different.

Marco laid out some key arguments:

Halos system: Unified full-stack safety platform integrating hardware, software, and tools.
Cosmos technology: Generative AI for creating diverse simulation scenarios and advanced reasoning models.
Dedicated chips & sensors: Improved hardware enabling faster, more reliable autonomous processing.
Internet pre-trained models: Leverage large-scale driving data for enhanced decision-making.
Simulation breakthroughs: Reduced simulation-to-reality gap, boosting training efficiency.

Here’s what’s new, and Marco’s big takeaways when it comes to the future of autonomous driving.

1. How Today's AI Finally Mimics Your Teenage Driving Lessons 🚘

Self-driving cars are ACTUALLY picking up passengers in San Francisco right now. No safety drivers. ZERO humans.

Not in five years. TODAY.

Marco explained the massive shift: autonomous vehicles now learn like humans do.

Previous approaches? Coding EVERY possible scenario. (Impossible!)

Now? AI brings lifetime knowledge to driving, just like you did at 16.

Foundation models give machines the same intuitive understanding you have. They know a cardboard box isn't dangerous but a ball rolling into the street means DANGER cuz there’s probably a little kid chasing it.

This isn't just fancy tech talk. Autonomous vehicles are literally driving paying customers around SF and other major cities as we type.

Try This:

This Bloomberg article takes a look at the autonomous car scene so far in San Francisco. The big takeaway?

Robotaxis spark convenience for some, chaos for cities.

2. NVIDIA's "Impossible Scenario Generator" Creates Dangers No Human Ever Imagined 🤯

Remember when testing meant driving millions of actual miles?

LAME.

Marco revealed how NVIDIA's Cosmos platform generates literally INFINITE dangerous driving scenarios no human ever thought to program.

Deer + construction zone + snow + dusk? Generate it. Then make 10,000 variations instantly.

Even wilder: they're using AI as JUDGES to evaluate millions of driving videos, filtering out the bad behaviors from training data.

Marco predicts this simulation revolution will transform the ENTIRE robotics industry beyond just cars.

Try This:

Ever wondered how endless, high-risk driving scenarios come to life without millions of miles on the road?

Explore Nvidia’s Cosmos platform—a breakthrough that leverages 20 million hours of video to generate synthetic, photorealistic simulations of complex driving conditions (think deer crossings in construction zones during snowy dusks).

3. Why an Italian Flashing Headlights Could Make or Break a Billion-Dollar Industry 🔦

In Northern Italy, flashing headlights means "go ahead!"

In Southern Italy? "DON'T YOU DARE cut in front of me!"

Marco dropped this mind-blower to explain why robotaxis may work in San Francisco but could struggle elsewhere.

Each region has tons of unwritten driving customs. Each new city requires expensive local training data.

The financial equation breaks unless cars can learn new regions faster with less data.

Marco's team is tackling this with a three-pronged approach: test fleet recordings, internet dashcam videos, and AI models that filter which behaviors to learn.

Their Halos stack unifies everything into one product that partners like Mercedes and GM are already adopting.

Try This:

Curious about how regional driving quirks can make or break autonomous vehicle success?

Check out Nvidia’s latest collaboration with General Motors. Their integrated Halos stack is helping automakers overcome the challenges that have plagued the autonomous vehicle industry for years.

⌚

Numbers to watch

$ 18 billion

Perplexity’s latest fundraising round has the company valued at $18 billion.

Reply

or to participate.