What if I told you that Anthropic is the
one that's actually ahead and that
Anthropic has had the better month than
OpenAI? I know that sounds insane
because they've had their model banned.
But, I'm going to explain why and I'm
also going to explain why maybe maybe
maybe none of it matters as much as we
think and we should be looking outside
the two major model makers for the
energy in the AI space. And so, I'm
going to cover a major new development
from not OpenAI and not Anthropic at the
end of the video. Stay tuned for that.
But first, let's jump into what made
OpenAI's week great and the news that we
didn't see about Anthropic that also
argues that they may be ahead. OpenAI
really looks like it won the last week,
right? All of the story has been around
how Fable has been banned, how there
have been unsuccessful negotiations with
Washington. Now, there are rumors maybe
it's getting back on track. Meanwhile,
OpenAI has rumors of 5.6 coming out and
also got a major hire from Google in
Noam Shazeer, one of the original
authors of Attention Is All You Need,
which is the paper that kicked off the
entire LLM revolution. So, one of the
original authors of that went to OpenAI.
It's easy to see, "Wow, it's a
king-making week." I want to tell a
different story. I want to suggest to
you that it's quite possible to read
this week and think about it and come
out with the opposite conclusion, that
Anthropic is the one that's ahead. Now,
I'm not saying I know that for sure, but
I'm saying it's worth considering and I
want you to sit here and think about it
with me because the implication of
having a wider field of view is that we
start to see the edges of what's
possible with these AI models. We start
to see beyond the obvious news. So,
stick with me. Anthropic
also got major talent this week. That
didn't make as much news, but John
Jumper went to Anthropic from Google.
Google had a terrible week. There's just
no other way to say it. Now, John Jumper
is a Nobel Prize winner in AI and he won
because of the work he did in chemistry
on AlphaFold with Demis Hassabis. And
now he's going to Anthropic. So, here's
the deeper story.
The talent movements matter much more
than the cover story with Fable being
banned temporarily and all of that. I
have every confidence that Fable's going
to come back. But, if you look at where
the talent is going, it's concentrating
not just at OpenAI, but at both major
labs, at Anthropic and at OpenAI. And in
both cases, the bet around the valley is
that recursive self-improvement is
starting to kick off at these labs. And
if that's the case, Anthropic is not
nearly as poorly off as the news is
making it seem this week because they're
ahead on recursive self-improvement. Do
you know how I know that? Because Fable
and Mythos, regardless of whether
they're banned,
are a new pre-trained model. There's
just no other way to interpret that.
Remember, Anthropic never invested in
the reasoning layer that OpenAI invested
in. A lot of the way that OpenAI turns
the crank on these models
is by improving the reasoning and
post-training pieces of the model,
improving some of the harness pieces,
how the model uses memory.
And that's how they release these models
that they call their point one model
releases. OpenAI, in fact, has not had
nearly the same cadence at pre-trained
models that are larger in scale as
Anthropic has because they've been able
to lean on reasoning. This is an
underappreciated point and it's about to
start to matter because pre-trained
models are naturally more intelligent.
Now, they're phenomenally expensive to
run because pre-trained models to scale
up, you have to spend more and more and
use more and more GPUs. But, Anthropic
has the edge right now because they have
the largest, freshest pre-trained model
in the world. And they can use that to
train the next one. And that is exactly
what I think that they want to do. Now,
OpenAI, of course, is working on a
pre-train as well, but keep in mind
their last major publicly disclosed
pre-train in 4.5 got pulled very soon
after it was launched. It did not go
super well. And so, one of the questions
that we've all had in the background is
when is the next major OpenAI pre-train
model going to come out? And if it comes
out, how is it going to be delivered,
how is it going to be put together with
the post-train and the reasoning side of
things to deliver a overall package
inside their codex, inside their ChatGPT
experiences? That is going to take some
time.
What I just described is not easy to do.
And so, I would argue that from a pure
model perspective right now, Anthropic
has the strongest position in the race.
Fable and Methuselah are the best models
in the world. They can use them to make
another jump in pre-training. And it may
be the case at this point in the race
that that one half step ahead that they
have is enough. I don't know that for
sure, and there's great talent on both
sides of the chessboard saying yes and
no. But, certainly, it it would be
irresponsible to count them out of the
race at this point. So, that was the
week for OpenAI and Anthropic. Now, the
part I promised you where I said that I
would cover the wider space. I would
argue that none of the conversation,
whether you agree with me that Anthropic
is ahead or not, none of it matters
compared to what Midjourney announced
this past week. Midjourney announced the
biggest medical imaging breakthrough in
50 years. They took their own business.
It's extremely profitable. They can
reinvest the money anywhere they want.
They have 40 employees and $200 million
in revenue. They got to put that money
somewhere. They decided to put it into
preventative medical care with a brand
new device. This device is a special
kind of ultrasound that images the
inside of your body and it's cheap and
it's affordable and it's much much
faster than an MRI and they're planning
to roll it out to the scale of getting a
billion of these scans a year in the
next couple of years. And it's going to
feel like a visit to the spa and that's
the way they're framing it. They're
going to build something that feels like
a spa in San Francisco first and then
they're going to expand from there. All
you do is you sit in this little pool of
water and you get this ultrasound
blasted at you for like 60 seconds, 2
minutes and then it's done. And it's
done. And then you have an image of your
whole body and because it's that fast,
it's more affordable. It can be done
preventatively. You can see
cardiovascular structures changing over
time. You can see other details in your
body changing over time which allows you
to pick up cancers earlier, allows you
to pick up heart disease earlier. And I
know that there are doctors who have
been concerned that this preemptive
imaging is going to be something that
gets people into false positives, right?
That they're going to identify all kinds
of things that aren't really wrong. But
we've never had imaging available at
scale before. We actually don't know
what that looks like. All of our imaging
technology is reactive by definition.
It's it's expensive. Not anymore. We're
going to have this cheap available
medical imaging that allows us at a
population level to start to understand
how our bodies change over time. It is
it is a breakthrough that is going to
save lives. It's a huge huge deal and I
think it's a great example of how energy
in the space is actually moving outside
of these two major players. I love
Midjourney as a product. Anyway, I've
talked about it before and I love that
they are choosing to bootstrap this.
This is an example of what you can do
when you build a successful business
model and you say, "You know what? I
think this is an important thing to do
for humanity and I'm just going to do
it." And I'm not beholden to it like a
VC board. I can just do what I what I
want to do here. And so, that's how the
innovation happened. They decided to
invest in it. The the founders have a
background in hardware and device tech
and they just went after it. I'm really
excited to see this one come out and I
will be talking more and more over the
course of the week about other ways that
we can be building and staying engaged
outside just this model horse race. But
if you want to talk about the model
horse race, I think the case is there
that Anthropic is ahead. So you tell me
what you think. Do you disagree? You
think OpenAI is ahead? Or do you think,
you know, Midjourney, the energy outside
the space is what matters? And then put
down what you think I should cover that
isn't OpenAI or Anthropic. Some of this
energy outside the space that would be
really compelling from a story
perspective that we should all should
pay attention to that tells us why AI
matters beyond just
who's ahead in a particular benchmark
score. I'd love to hear about it.
Cheers.