😸 Shocking Waymo safety stats

PLUS: OpenAI's eye-popping new valuation.

Welcome, humans.

Every now and then we come across great AI cat content. Please enjoy this recap:

Cats in space (the transitions in this video are great):

Cats surfing (love this art style):

Cats helping with the dishes:

And this non-AI cat that appeared IRL while we were writing today’s newsletter!

V proud of this action shot

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

Waymo claimed self-driving cars outperformed human drivers.
Mistral launched Pixtral 12B, an open-source multimodal model.
OpenAI plans to raise $11.5B at $150B valuation.
U.S. senators suggested AI-summaries could violate antitrust laws.

Advertise in The Neuron here.

Self driving cars = better at driving than self driving humans.

Actual footage of Pete and Noah in a Waymo last July.

If you haven’t already heard, driverless cars are all over San Francisco these days.

And Phoenix. And Austin. And Miami. And Seattle. And Los Angeles. Oh, and did we mention Atlanta?

By most accounts, Waymo (owned by Alphabet) is the company to beat in self-driving:

It has a ~700 car fleet.
It now sells 100K rides per week (up from 10K a week last year).
Waymo cars have driven 22M miles over 2M paid rides.

After a bumpy launch last November, where self-driving cars were causing all kinds of impediments around SF, Waymo now says its autonomous vehicles cause fewer than one crashes per every million miles driven.

In 22M miles driven, Waymo has only reported:

20 crashes with injuries.
5 crashes serious enough for an airbag.
1 crash involving serious injuries.

Waymo says in the same amount of miles, human drivers in SF and Phoenix would have caused 64 crashes, 31 of which would have triggered airbags.

If you’re mind-blown by those facts, same.

Now get this. Timothy B Lee of Understanding AI dug deep into this data, and found some fascinating facts:

Out of the 23 most serious Waymo crashes, 16 were because a human rear-ended a Waymo car.

And 3 more involved a human running a red light and hitting a Waymo (like this one, where the Waymo was a victim of the crash, but still had to report the accident).

Let’s do some back-of-the-napkin math: 16+3 = 19, 23 - 19 = 4, so that means only 4 serious crashes were actually a Waymo’s fault.

By Timothy’s analysis, Waymo’s injury crash rates are approximately 60-70% less than human drivers. Sounds like the worst part about self-driving cars is the self-driving humans you have to share the road with!

Our take: The safest self-driving car tech will be the one everyone uses—we might not be the safest drivers, but us humans don’t mess around when it comes to “unsafe” new tech.

Check this report from S&P Global, which now projects there will be 230K driverless taxis sold in the U.S. by 2034—wonder how many of those will be Waymos?

What did you think of today's main story?

🐾 I prefer purely AI content

🐾🐾🐾 No better or worse than usual

🐾🐾🐾🐾🐾 Loved it — thanks for covering new emerging tech!

FROM OUR PARTNERS

Attention is your secret weapon for automated email follow-ups.

In sales, timing is everything. But how can you stay on top of every lead when you're juggling dozens of calls a day?

Easy—let Attention's AI assistants help you:

Generate personalized emails in under 60 seconds.
Automatically update your CRM with call insights.
Receive real-time coaching during calls to nail your pitch.

Don't let crucial details slip through the cracks. Let Attention handle the heavy lifting so you can focus on what you do best—closing deals.

If you want to automate your email follow-ups and close more deals, check out Attention here.

Around the Horn.

Adobe Firefly Video Model Coming Soon | Adobe Video

Mistral launched Pixtral 12B, its first open-source, multimodal model (Github here), with testable versions coming soon.
OpenAI will raise $6.5B in equity and potentially $5B in debt at an $150B valuation (up from $86B in its last round).
Amazon will test ads in its Rufus chatbot, and will be based on conversational context and Amazon search data—Rufus may even generate text alongside ad copy as well.
A group of U.S. senators now believe AI-generated summaries could qualify as an “antitrust violation”—in very related news, Google is now in the middle of its third antitrust trial in the span of a few years.

Treats To Try.

*If you’re new to AI or need the confidence to launch your product, you need Vellum’s free GenAI webinar. Learn best practices from AI experts at major companies like Redfin. Seats are limited—grab yours now!
Glean is a work platform that helps employees find info, create content, and automate tasks from internal knowledge.
FiveThirtyNine claims to be better than humans at predicting the future (and it’s a pretty good research tool)—try it here.
CodeViz creates interactive visual maps of codebases to help developers understand and navigate complex software projects.
StoryArtAI generates unique illustrations for children's books (here are their tips on consistent characters).
Invoke creates “ethical” art for video game studios.
Llama-Omni provides real-time speech interaction with Meta’s Llama 3.1 8B, to generate both text and speech at the same time. Get the code here.
Verse is a new creative app that lets you make mini webpages that double as mood boards, event invitations, storefronts, or loads more.