Welcome to Crash News.
Today is May 13, 2025.
And speaking of the future,
getting here maybe faster than we thought,
did you catch wind of this sort of quiet shift happening
in AI infrastructure?
- Oh yeah, what's the buzz?
- Well, it seems like it's not just about
who has the biggest, smartest model anymore.
It's becoming more about smarter strategies
and maybe some unexpected players calling the shots.
- Interesting.
- Yeah, and here at Crash News,
we try to be your shortcut to understanding this stuff,
the really important AI news.
We take all the complex things
and break it down step by step, plain English.
- Starting simple, then building up.
- Exactly.
We'll start with the basics, layer in more detail,
and always pause if there's a term that might be new.
Explain it quickly.
- Makes sense.
- Cool.
So before we jump in, just a quick reminder,
if you find this helpful,
there are some totally free ways you can support Crash News.
A quick rating, maybe a like, subscribe,
on whatever platform you're using to listen,
that helps a lot.
- Definitely does.
- You can also hop into our Discord community.
It's going pretty fast, lots of good chats there.
And for anyone who wants to do a bit more,
we do have a Patreon.
- Great options.
So what are we diving into today?
- Okay, three main things.
First, OpenAI's really ambitious Stargate data center plan.
Sounds like it might be hitting some turbulence.
- Turbulence, okay.
- Then we'll look at Google.
They're making a strategic move,
funding AI startups with a new fund.
- Ah, fostering the ecosystem.
- Exactly.
And finally, we'll see how different fields
think high stakes, like healthcare and journalism,
how they're actually dealing with bringing AI
into their day to day.
- Right, the practical.
side of things. Okay, let's start with OpenAI and Stargate. What's the story?
Okay, so the main headline is this huge initiative, Stargate, a $500 million data
center project. It's reportedly, well, facing some difficulties. Difficulties like. Primarily,
it seems to boil down to investor caution, you know, linked to the general economic volatility
we're seeing. Okay, so before we go further data center, for anyone maybe picturing, I don't know,
a closet full of servers from a movie, what are we really talking about here? Right, good point.
Think bigger, much bigger. Imagine like a massive warehouse, just absolutely packed with rows and
rows of incredibly powerful computers. These aren't your desktop PCs, they're specialized
machines. And AI models need these for, well, two big things. The learning phase, which we call
training, that takes immense power, and then just running the AI, answering queries, doing tasks.
Got it. The heavy lifting hardware. Exactly. So Stargate is meant to be a particularly enormous
and powerful version of this, specifically for open AI's future, potentially much larger AI models.
Okay, so the people with the money, the potential investors, are feeling a bit nervous about the
economy right now. Makes sense that a half billion dollar project might seem riskier.
That seems to be the core issue, yeah. Hesitancy to commit that kind of capital in uncertain times.
But here's where it gets interesting, maybe a bit counterintuitive. While open AI might be facing
delays, the situation in China looks, well, quite different. Yeah, that's a fascinating contrast,
because reports suggest China actually has significant overcapacity in its AI
data centers already. Overcapacity, like empty space.
pretty much we're hearing figures like maybe 80% of the resources across over
500 recently built data centers are just sitting idle Wow 80% that's the report
and this is even after you know market adjustments happened because of new
maybe more efficient models like deep seeks are one which changed what kind of
hardware people needed so even with all that unused capacity they're still
investing heavily it seems so there's a really strong push to continue building
out this infrastructure likely to compete directly with u.s. initiatives
like Stargate okay so let's piece this together a major u.s. project Stargate
might be slowing down because investors are cautious about the economy right
meanwhile China despite already having tons of unused AI computing power is
doubling down on building more exactly it kind of paints a complex global
picture doesn't it it really does and it raises a big question maybe for you
listening if a project like Stargate slows down what does that mean for how
fast AI develops overall and maybe just as important who ends up controlling the
fundamental infrastructure these massive data centers that all this powerful AI
relies on yeah definitely something to chew on okay let's shift gears second
topic yeah Google they've launched something called the AI futures fund
what's that about right the AI futures fund this is Google essentially setting
up a program to invest in early stage startups the key thing is these startups
have to be using AI tools and models from Google DeepMind to build their
products so it's not just money then no it's more of a package deal besides
potential funding these startups get early access
to Google's newest AI models.
They get credits to use Google Cloud.
- Cloud credits are always useful for a startup.
- Absolutely.
Plus they get support directly from Google's AI experts,
researchers, engineers, and yeah,
there's also the chance of Google
taking an equity stake directly.
- Sounds like a pretty sweet deal for a new company.
Do we know who's involved already?
Any early participants?
- Yeah, they've mentioned a few.
There's Viggle, they're working on AI for video creation.
Synthesia is another one focused on synthetic media,
like AI avatars, and Replit,
which is a platform for collaborative coding.
So you see, it's a pretty diverse mix of applications
they're looking to support.
- Right, so obviously it helps the startups.
But what's Google's angle here?
What's the bigger strategy?
- Well, if you zoom out, it looks pretty strategic.
By supporting and investing in companies
that build on their tech,
Google is essentially trying to make its AI models
the go-to standard.
- Ah, like building an ecosystem around their tools.
- Exactly that, cultivate this environment
where new businesses, new developers
just naturally choose Google's AI platform
to build on top of.
- Makes sense.
And speaking of developers,
Google's also jumping into the AI coding assistant race,
right?
- That's the other big piece of news, yeah.
The word is they're planning to unveil their own AI agent
for software development,
likely at their big IO conference May 20th.
- Okay, so an AI tool to help write code, debug,
that sort of thing.
- Presumably, yes.
- Assisting developers through various parts
of the software creation life cycle.
And this is significant because it puts them head to head
with existing players.
- Like who?
- Well, you've got anthropics clod code,
open AI has some.
something reportedly called Windsurf.
And then there's a whole bunch of smaller startups trying to crack this space too.
So that area is definitely heating up?
For sure.
It's seen as a major application area for AI.
And I heard something about AR glasses integration.
That sounds pretty futuristic.
Yeah, there's some speculation floating around about that, that this new coding assistant
might integrate with Google's Gemini AI models, and maybe even their augmented reality hardware.
Wow.
Imagine coding help projected right into your glasses.
Exactly.
So taking these two things together, the Startup Fund and this new coding assistant, it leads
to another question for you.
How are these Google moves going to change the AI innovation landscape?
And maybe more personally, what kinds of new opportunities could this open up for people
wanting to start businesses or just for individual developers?
Lots to think about there.
Okay, let's move to our third area.
How AI is actually being used or maybe not used in different industries.
Let's start with healthcare.
Open AI introduced something called Healthbench.
Yes, Healthbench.
This is a new benchmark Open AI developed specifically to test how well AI models can
handle real, often complex medical conversations.
Okay, so like a standardized test for medical AI chatbots?
Sort of, yeah.
But what's really important is how it was developed.
They got input from 262 doctors across 60 different countries.
Wow, that's a lot of expert input.
It really is.
Healthbench's benchmark is hopefully grounded in, you know, actual real world medical situations
and the nuances involved, not just textbook cases.
So what does it actually measure?
How does it test the AI?
The AIs are evaluated on how they respond.
in over 5,000 genuine health related conversation scenarios.
The physicians helped create a detailed scoring system,
a rubric, looking at key things like,
is the AI factually accurate?
How good is the communication?
Is it clear, empathetic?
And crucially, does the AI know
when it needs more information?
Does it ask the right context seeking questions?
This was tested across all sorts of scenarios
from like emergency referrals
to broader global health issues.
- Right, knowing what you don't know
is pretty critical in medicine.
- Yeah, absolutely vital.
- So what did they find when they ran different AIs
through this health bench, any surprises?
- Well, one interesting finding
was that OpenAI's own O3 model
apparently did better on this specific benchmark
than other big names like Clod 3.7 or Gemini 2.5 Pro.
- Okay, not entirely expected
they'd highlight their own model doing well.
- True, but perhaps more surprising
and potentially more significant
was the performance of their smaller models,
something called GPT 4.1 Nano
showed really strong results.
It actually outperformed some larger models,
even their own GPT 4.0 on some aspects
while being way cheaper to run.
- Cheaper how?
- The report mentioned something like 25 times cheaper.
- 25 times, whoa.
- Yeah, so it suggests that maybe for specialized tasks
like these medical conversations,
you don't always need the absolute biggest,
most expensive AI model to get top tier performance.
- That's actually a huge deal for potentially
making healthcare AI more accessible, more affordable.
- It could be, yeah, it opens up possibilities.
- Okay, fascinating stuff in healthcare.
Now let's switch gears completely.
Journalism, how are newsrooms dealing with AI?
Are they embracing?
it, resisting it. Well, the picture there seems much more nuanced, cautious is probably a good
word. There was a survey recently talking to people at places like Raiders, The Washington Post,
VentureBeat, 404 Media. Okay, reputable sources. Right, and it shows newsrooms are being very
selective about where they use AI. There seems to be general acceptance, even enthusiasm,
for using AI tools for things like transcription turning audio or video into text. That saves a lot
of time. Sure, I can see that. Also for data analysis, like sifting through large data sets
for investigative pieces, and for translation, making stories accessible in other languages.
So tools that help the journalist do their job more efficiently. Exactly. These are seen as
enhancing efficiency or reach, but they don't really touch the core process of creating the
actual news content. So that's where the line is drawn, actually writing the articles. Precisely.
There's a clear reluctance, even resistance, to using AI for that primary content creation task.
The journalists surveyed really emphasized things like maintaining audience trust,
upholding journalistic integrity. Makes sense. You need to trust where your news comes from.
Absolutely. There's a big worry that AI generated content just might not meet those standards or
could be easily misused. For example, Reuters mentioned they use AI to help generate about
25% of their code now for internal software and tools. Okay, so using it for technical tasks
internally. Right, but they remain very skeptical, very wary about having AI write the actual news
stories that people read. So it's more about AI as a supportive tool, augmenting the human
human journalists, not replacing them.
That seems to be the dominant approach right now.
Careful integration.
OK.
So when you look at these two fields, health care and journalism, with their different
approaches.
Yeah.
It really highlights some key things, doesn't it?
These practical considerations that come up as AI gets woven more deeply into professional
life.
Like, how do you guarantee accuracy and safety in a field like medicine where mistakes can
be critical?
Right.
And in a field like journalism, how do you protect fundamental values like trust and
integrity when the technology could potentially undermine them?
These are big ongoing questions.
They really are important things to keep wrestling with.
Well, I think that brings us towards the end of our deep dive today.
Yeah.
So just to recap quickly, we saw how even huge AI projects like Stargate can bump up
against economic reality.
Right.
The money side matters.
We saw how tech giants like Google are playing a strategic game, trying to shape the future
by funding startups and building new developer tools.
Building that ecosystem.
And we saw how different industries, health care, journalism, are finding their own paths,
cautiously integrating AI, balancing the potential benefits with their core values and concerns.
So hopefully, stepping through all that has given you a clearer picture of some key things
happening right now in AI.
Maybe a few aha moments without it being too overwhelming.
Yeah.
And maybe leaves you with a final thought to ponder.
As AI keeps getting more powerful, more integrated into everything, what role should critical
thinking and just a really nuanced understanding of what AI can and can't do?
What role should that play in how we develop it and adopt it everywhere?
Hmm.
Good question to leave folks with.
Thanks for tuning into this deep dive into the world of AI.