AI Surge: $100B Fund, Cost Debate, O1 Preview
Key Points
- Microsoft and BlackRock announced a $100 billion AI fund, signaling confidence that the AI boom is far from peaking and betting on massive training infrastructure for the mid‑to‑late 2020s.
- A Washington Post piece on AI energy use was challenged by a senior tech policy fellow who calculated the cost of a GPT‑3 call to be about 2 cents—roughly 370 times cheaper than the Post’s estimate—highlighting the need for accurate cost reporting.
- Accurate cost metrics are crucial for meaningful debates about AI’s expense versus the opportunity cost of human labor, such as comparing an AI‑generated email to one written by a person.
- The newly launched 01 Preview platform is already producing functional apps, exemplified by a sleek weather app that only required plugging in a public API, sparking interest in building similar tools.
- The speaker teases an upcoming Maven course that will dive deeper into these topics and guide listeners on leveraging AI developments.
Full Transcript
# AI Surge: $100B Fund, Cost Debate, O1 Preview **Source:** [https://www.youtube.com/watch?v=rnDbLdW5wMI](https://www.youtube.com/watch?v=rnDbLdW5wMI) **Duration:** 00:09:06 ## Summary - Microsoft and BlackRock announced a $100 billion AI fund, signaling confidence that the AI boom is far from peaking and betting on massive training infrastructure for the mid‑to‑late 2020s. - A Washington Post piece on AI energy use was challenged by a senior tech policy fellow who calculated the cost of a GPT‑3 call to be about 2 cents—roughly 370 times cheaper than the Post’s estimate—highlighting the need for accurate cost reporting. - Accurate cost metrics are crucial for meaningful debates about AI’s expense versus the opportunity cost of human labor, such as comparing an AI‑generated email to one written by a person. - The newly launched 01 Preview platform is already producing functional apps, exemplified by a sleek weather app that only required plugging in a public API, sparking interest in building similar tools. - The speaker teases an upcoming Maven course that will dive deeper into these topics and guide listeners on leveraging AI developments. ## Sections - [00:00:00](https://www.youtube.com/watch?v=rnDbLdW5wMI&t=0s) **AI Billion‑Dollar Bet & Energy Controversy** - The speaker surveys recent AI news—highlighting Microsoft and BlackRock’s $100 billion AI fund, a disputed Washington Post report on AI energy consumption, and a teaser for an upcoming Maven course—arguing that the AI boom shows no signs of peaking. ## Full Transcript
you know the last few days in AI have
been absolutely wild I just want to run
through a few of the things that have
happened and a little bit of broader
economic news and I want to talk to you
about my Maven course that's coming so
stay tuned there's a bunch to get
through first thing is Microsoft and
black rock have launched a$ 100 billion
fund for AI so if you think the AI
bubble is at its peak if you think it's
a bubble if you think it's going to pop
it's not at its peak yet if they're
launching hundred billion dollar funds
they're expecting to get to a super
cluster or some kind of massive training
center in the mid to late 2020s maybe
2027 2028 it may be in time for the
training run that gets to artificial
general intelligence depending on what
you think about that depending on your
timelines they don't know either by the
way so that's why this is all very
uncertain they are taking a bet they're
taking a hundred billion dollar bet that
it is worth getting into that market so
that's number one number two the
Washington Post published a big article
on the energy consumption of AI and it
may not be as accurate as they say uh
and the reason I call that out is that
there was a very widely cited uh post on
X or the site formerly known as Twitter
uh that basically took apart the
components of a llm call and came out
with a very different
number it was 370 times less expensive
than the number cited by Washington Post
now you can say hey it's X I don't buy
it this guy is a senior policy fellow in
Tech and what would he know but even if
you don't buy it the wo article cost
actually came out to around 2 cents per
call and that's cheaper than a letter
and we don't complain about letters
being
expensive we don't now not a lot of us
send letters anymore because emails are
easier but still the point
being if we're going to talk about the
costs of AI we need to talk about them
accurately in terms of the real costs
involved we should not site gpt3 which
is what the Washington Post article did
and we need to be really clear about the
opportunity cost as well what is the
opportunity cost of a human writing an
email versus an AI writing an email that
was one of the prominent examples in the
post so more to come there but I wanted
to flag that because there's been a lot
of comments I've seen coming in
basically saying AI is really expensive
and I think articles like the post need
to be accurate in their report of cost
so that we can have a real conversation
about
this um and then next I wanted to call
out 01 preview so 01 preview has been in
the news recently I've talked about it
on this channel briefly we have now had
it for a few days we are now getting
apps built by 01 preview frankly I saw a
weather app built by 01 preview that
looks nicer than the weather app on my
phone all it needed was an API to be
functional and those apis for weather
are ubiquitous you can just plug them in
and go makes me want to build a weather
app for my phone maybe I should do
that but that's an example of how
quickly it can sort of put together a
fully fledged application in just one or
two prompts one of the things I've
noticed with o1 is that 01 is really
really helpful at debugging code and I'm
not the only one so I gave it some code
written by Sonet two nights ago it took
30 seconds to think about it and it came
back with a much cleaner structure
it successfully pulled Sonet out of a
death spiral with the code and I got the
code
working I heard other people uh that I
know of talking about uh 01 debugging
3,000 lines of code and finding a single
character out of
place so that matches my anecdotal
experience it's really good at debugging
the other thing that I want to call out
it's not just code right people are
identifying it as a good editor and a
good editor is different from a good
composer when you write with sonnet
sonnet tends to overwrite it tends to
write an entire text and then if you
tell it to change something it just
rewrites the whole text whereas 01 can
be more nuanced about the changes it can
it can use a a scalpel right it can
precisely adjust things so that they
sound
better we're still at the beginning of
discovering what 01 can do but I think
that the way I'll leave this the
challenge I have for you is if you don't
think o1 can do something if you don't
think a large language model can do
something why have you tried it if you
say oh it's the inputs oh you know 01
won't accept images1 won't accept
whatever it is there's ways around that
you can take the text and you can stick
it into a prompt and now you have the
full text of the document and the prompt
just as an example you can also switch
models partway through the chat you can
be talking to chat
gp40 and then partway through the chat
switch over to 01 and see what
happens another creative idea that I saw
H and came across that I really love is
pick a task that you don't love record
yourself doing it in a loom video and
talk your way through it then take the
transcript and upload that to an llm and
ask it what can be
automated just see what can be automated
I would be really
curious I bet it has some ideas and I
bet it could help you automate so we're
at the stage where my bias is to ask the
llm and I think that's a huge difference
you can actually see it in the chatbot
Arena scores so chap out arena for those
who don't know is a
gigantic um well that's what it sounds
like it's an arena where people compete
against uh basically rank uh various
llms against each other to see what the
quality of answers are and it's
crowdsourced so like no given company
can really game
it and the problems are crowdsourced so
again nobody can really game
it and I will tell you it is an
absolutely jaw-dropping graph when you
look at 01 and mini and 01 preview they
are like 100 points better uh than any
other model and by the way we're talking
100 points is a lot so everybody every
model right now is within the same 50 to
70 Five Point radius somewhere around
the 1250 Mark to 1200 roughly in Elo or
ELO
ratings which is basically a
mathematical way of estimating relative
strength versus another player it's used
in chess a lot actually but now it's
used you know to estimate chatbot
competency this is for
mathematics 01 and 01 pre or 0 preview
and 0 mini are over 1350 they're much
much better a step change better than
other chatbots and I call that out
because there's been a lot of talk about
mathematical reasoning and the ability
of a model to do reasoning this is just
the preview Sam Altman has called out
that reasoning is about to get better I
think he described it as the g at the
gpt2 stage with 01 preview he thinks
reasoning can get immensely
better and he says that the full 01
model is coming in just a few months so
if you think well there's weaknesses in
01 I don't know well just just wait
right just wait a couple months uh
you'll you'll be surprised
Okay so we've talked about 01 we've
talked about some of the AI news I want
to talk about economic news the FED cut
rates by half a point that is a big
deal it means that there is more likely
to be capital in the system for Tech and
you thought the hundred billion doll
fund was Capital there's more likely to
be Capital available for startups as a
whole not just AI startups there's more
likely to be an appetite for hiring it's
good for jobs people are more likely to
be confident in the economy housing
prices are going to get a little boost
because mortgages are cheaper to
purchase it's good for everybody we've
been waiting for this rate cut a long
time there may be a couple more coming
later this year we will see they cut
point they cut the rates by more than
expected by the way the expectation was
just a quarter of a point so they
doubled that to half a point we'll see
where this lands but it's very very good
news for those of us in Tech who have
been waiting a long long time for rate
cuts to loosen capital in the
space and last but not least I want to
call out that my Maven course is live
folks have already enrolled I am excited
to launch it if you are interested in
signing up in getting on the wait list
in enrolling you can check out the link
I'll post below and I will post a
special discount code for you in the
chat
underneath there you go for what it's
worth
uh and that's what I got please enjoy