June 5, 2023 • 32 mins
We discuss a number of toping relating to markets and society.
- Apple’s Savings Account and is it a new Financial Player
- Future of retail banking
- Can payments ever be costless?
- The weight and cost of financial regulation
- AGI and open source AI tools lead the way
- The power of an independent AI developer scene
- What does AI mean for companies large and small
Andrej Karpathy: Tesla AI, Self-Driving, Optimus, Aliens, and AGI | Lex Fridman Podcast
https://www.youtube.com/watch?v=cdiD-9MMpb0
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
hello.
Good evening, good afternoon,good morning, wherever you are.
This is Off the Fence, atransatlantic podcast where
three guys in Madrid and in theNew York metropolitan area get
together and try to put theworld to this week we're gonna
talk about developments inbanking, in US banking and
specifically Apple offering asavings account.
(00:21):
It's gonna actually pay acompetitive interest rate which
should be interesting given theway deposits have been moving in
the US and and other and otherdevelopments.
But Luis, this was your topic.
Why didn't you, why don't youtake it, tell us what you think
what the issues are.
Thank you Peter.
I think, we discussed this topicin previous podcasts and.
(00:44):
My impression is that whenyou're presented with such a
fantastic spread betweendeposits that are site deposit
that pay nothing and federalconcentrated 5% or 500 basis
points spread between what bankspay for deposits and treasuries
short data treasuries, therewill be somebody somewhere that
(01:06):
will take advantage of that totake market share in deposits
and do something with it.
And we have, we, we talked aboutit and we couldn't, I couldn't
figure out who would be thedisruptor.
Apple yesterday broke the ice.
It is, in my opinion, a muchmore significant attacker than
any of the attackers that we sawin the early 2000, such as I n g
(01:28):
Direct or the online brokers.
Various reasons.
First of all, apple as an, asthe.
Owner of the operating system inwhich a lot of people's lives
work has an enormous amount ofsubscribers already many of whom
use different functions that theApple Pay app allows them to
(01:51):
use.
It is also a big contender inthe buy now pay later layaway
system of buying for retailpurchases.
And therefore the fact that theycan get funding at a competitive
rate for them and maybe not forthem, maybe somebody else who
does the actual financialmanagement of both the asset and
(02:12):
the liabilities of these in, ofthis book.
I think it's quite a threat tosome lenders to consu to some
consumer lenders in particular.
Unsecured consumer loans.
One has to believe that it won'tbe too long before Amazon thinks
that maybe this is also theirdomain, that they should
(02:35):
probably get into this.
And perhaps other of theplatforms that are involved in
either payments or consumerRetail might, will also develop
their own special, theinteresting savings products
with this.
I think the one belonged beforethis trend moves from west to
(02:57):
east as these trends generallydo.
And discuss with you guys inprevious podcast.
I believe that the currentexcitement with cyclical
Companies, specifically banksin, in the Eurozone might be a
little bit ahead of itselfbecause, this would be one of
the threats that we could thinkof for that markets.
(03:20):
What do you think?
I, go ahead Alex, please.
No, please, Peter.
I'm having a look on onlineright now.
What big banks.
Our paying is an average rate,and it's about 0.1%.
And most if you look acrosssmaller banks, they're, they get
to around 3%.
(03:41):
I see you have some specialoffers.
PNC has an account at 4%.
But this is a competitive move.
Banks are slow to raise depositrates.
We know that they're a lotquicker to raise their lending
rates.
It's which is natural.
Goldman Sachs, who will beservicing this, took a stab at
(04:02):
retail banking in a, with a unitcalled Marcus that hasn't gone
so well.
It hasn't been unwound, but it'sbeen it's been reduced that
back.
Apple and Goldman have beenworking together for a while.
And especially on the card side.
I, and, but this is a new andinteresting move.
It's interesting also causeApple has none of the capital or
(04:25):
I believe liquidity concernsthat other banks might have in,
in backing up a savings account.
And this has become an issue inthe last few, we, few months as
because Yeah.
What sovereign Silicon ValleyBank was was liquidity.
The liquidity can moveincredibly rapidly.
And this is a huge concern forregulators and supervisors.
(04:47):
Supervisors right now, I don'tthink, I don't think Apple
offers a deposit cause Appledoesn't own a bank.
I think the deposit have to lookinto the small print is probably
deposit at Goldman Sachs or somethird party bank.
Apple just is the facilitator.
I think.
The marketer.
Yeah.
But that's the whole point,right?
So few things.
That's right.
So the re what I'm trying tosay, Alex, is that the
(05:09):
regulatory constraints on thedeposits will be with the
department taking bank will bewith Goldman.
That, that's the whole question.
The funny part is that Marcusoffers today 3.9% versus four
and change percent.
You're not gonna your ownproduct are you?
Yeah, of course.
Exactly.
You're gonna love that.
And it shows you the power ofthe thing.
So a few things you mentioned atthe beginning, the impetus for
(05:32):
this and it makes sense is, thefact that the spreads have widen
significantly and that the banksbeing slow to raise rates, as
Peter mentioned.
Given that we've had, a good 10years of customers being used to
not earning anything in theirdeposit accounts.
This is, marketing wise kind ofan important move for a.
National player as opposed to aregional player.
(05:52):
Regional recently certainly havehad to fight for the deposits by
paying higher rates.
It's a move that makes a wholelot of sense.
And then actually like similarto SVB in some ways, what you
worry about is with an inflow ofsignificant amount of savings in
all one swoop in a fairly bigmarketing push by a very
important player that has a lotof customers, you really wonder
(06:13):
where, where those deposits aregoing to go and who's taking the
gap book decision.
You'd hope that it's not Apple,as you mentioned, since they
don't have expertise,credibility, or even a structure
to do it.
So as you say, most likelysomebody else is doing it.
And if it's Goldman, fantastic.
It's Goldman Sachs, I mean atthe Apple.
There you go.
(06:33):
Starting today, apple Card userscan choose to grow their daily
cash rewards with savingsaccount from Goldman Sachs,
which offers high yield annualpercent in yield, annual
percentage yield of four point15% rate.
That's one than 10 times thenational average.
No fees, no minimum deposits.
Yeah.
It's a perfect marketing thing,but the argument that I'd like
to make, just to move theconversation elsewhere is that
(06:54):
for sure we have an opportunityright now because we've had
this, the rates back up andexpectations have been very low
and it takes a while forcustomers.
Customers will look at this asbeing a good deal.
That's great, but my argument isa different one, which is that
Apple should.
Itself, long term B, in thetransaction processing business,
because one of the main factorsthat keeps the wholesale rate of
(07:16):
transaction processing in the USat 2% right is fraud.
And if there's one thing thatthe marriage of software and
hardware in your hand, has beenable to handle is biometric
verification and really gettingan end-to-end between the
receiver and the sender of a ifnot trusted, but.
Verified identity ecosystem,which is quite strong.
(07:41):
The apple has both the physicalhardware and the software to run
transactions.
Again, forget crypto, forget,how it gets handled.
But the idea is for them to be apayment processor, given that
they can ensure what is probablythe largest piece.
Of the puzzle in terms of thecost of wholesaling, transaction
processing I think long termease and attractive is an
(08:03):
attractive business.
It could very well be that theywanna grow their marketing shops
at this point, which makessense, and turn that into over
time, building a real financialbusiness where they do
transaction process.
Not take, necessarily back andbalance sheet risk, but just be
there to make sure that you havethe.
Funds in your account and thatyou're making an accurate
payment.
That to me, I think is a longterm play That would be a
(08:25):
significant source of growth forApple.
And what does it mean for banks?
I don't know if you can hear me.
Can you hear me?
We can.
Yes.
And just because of that, I'mgonna answer it.
The the problem with banks, Ithink is that the physical
existence of a bank branch mayhave outlived its relevance.
(08:46):
And as such, you have a lot oflegacy costs associated in a
cost structure that makes itfairly difficult to, to compete.
Those kind of banks are gonnabe, this is not a new theme.
People have been talking aboutit for 20 years and it hasn't
happened, of course.
But I think more and more mysense is particularly as you get
to alternative ways ofprocessing transaction, which
are almost costless, it's gonnabe very difficult for people to
(09:10):
accept the fees associated withtraditional banking.
That's, and that I agreecompletely.
And the, and place where feesare most obvious is in payments.
Yep.
And for years, banks basicallyhad a monopoly on payments.
They no longer do.
Yet there's still competitive,there's still room for new
competition in payments and andApple is part of it.
(09:34):
Others are in the field and andbanks are feeling it and bank,
especially in Europe.
Agreed, yep.
I think if I may say somethingvery obvious the big difference
between.
A bank like i n g direct that inthe early nineties started
collecting deposits by paying acompetitive rate and then didn't
(09:56):
know what to do with thosedeposits.
And the current situation isthat Apple also originates an
asset product, which is the,these layaway loans.
And they are the platform whereyou can do both.
Whoever they partner with thatruns the both portfolios.
We'll be very keen to see that,this works as, as well as it
(10:19):
does into the advantage of Appleand the financial partner they
have.
And I think that if theycontinue on this vein and their
opportunity for other products,it's probably different in
various parts of the world, but,I am going to go to meet some
people from Apple because, thereis an enormous opportunity for a
(10:40):
brand like Apple to develop amass affluent independent
financial advisor business inEurope, which is probably one of
the areas of the market wherethere's the most fat in the
world.
Yes.
And and this would be terrificfor the vast majority of the.
Population of the European Unionor and other Europeans.
But what I would also remind youof is that a few years ago,
(11:04):
Michael Milken, in one of hisappearances at one of the
conferences with, asked toprovide advice to some of these
platforms.
And what he told them is, don'tget into finance because the
regulation will bug you down.
I talked this morning I shouldcorrect myself.
I exchanged WhatsApps thismorning with one of the most
(11:27):
senior people that I have accessto WhatsApp at eight o'clock in
the morning in finance.
He was the chairman of a bank inSpain.
He was the vice chairman of oneof the top four investment banks
for Europe.
And I asked him what, where hethoughts about this, and one of
the things he said is, thiscompany was doing very well, UN,
(11:47):
until or has done very well.
Let's hope they don't want toget into the MOAs of being
regulated as a financial entity.
And I, and my, what I understandis great about this deal with
Goldman Sachs is that GoldmanSachs already pays all the fixed
costs of being regulated as afinancial entity and if they can
get the additional business, Ofbeing Apple's joint venture
(12:11):
partner in this, everybody'sbetter off and including the
public probably.
So I think it's a, it's anincredible, incredibly I think
it is as material to the bankingto the retail banking world as
the cash management account in1973 was to the retail banking
(12:32):
world when Marin h introducedthat.
I think you're right.
And there's something else aboutit, which is in recent years I
have heard an argument inbanking, which goes something
like this.
A bank does three things.
It takes deposits, it lendsmoney, and it makes payments.
It affects payments.
The payments business it's goingit's, if it's not gone, it's
(12:55):
going.
Others have entered.
You don't really need a bankinglicense to do it.
They're companies that can do itwithout the legacy.
The lending business is also isalso going private equity and
others are now lend money.
Amazon lends money to small andmedium enterprises and
securitize to to fund it.
(13:16):
There's a lot going on in thelending space.
Buy now, pay later is done by anon-bank, bank financial
institution or can be done.
So there's a lot going on inthat space.
But deposits, taking deposits iswhere the regulation kicks in
when you're a deposit takinginstitution, and thus others
weren't gonna go into it.
(13:36):
The, they may try to createAnna, the instruments that look
like deposits, feel likedeposits, but they're not gonna
be able to offer customersguaranteed deposits because they
don't wanna be regulated.
As Luis has just pointed out.
This seems to be a way in.
You, this partnership withGoldman Sachs to to get a big
tech into that space, into thedeposit taking space.
(13:59):
And these will be depositsbecause I'm, they'll be
guaranteed by the F D I C.
So I, I agree.
It's, it this is potentially abig move.
I agree.
I think what's interesting abouthere is you have two best of
breeds, honestly, in, in mymind, between Apple being
knowledgeable on the softwarehardware side, consumer loyalty,
(14:20):
privacy brand and obviouslyGoldman Sachs on probably
regulatory management and alldifferent financial
capabilities.
So you have two best of breeds,which for the moment early are,
have compatible.
And corresponding strengths intrying to address this.
My, my question would be longterm, I think to both your
(14:43):
points, how forward integratedinto this, does Zapp want to be
long term?
Right now it's easy.
It's a marketing deal.
That's no problem.
Longer term, it's gonna be abusiness for them.
But if it is, it's a completelydifferent business than what
they do and would be, enormouslyrisky.
And I think for the investmentbase, for Warren Buffet, for
everybody else around would beviewed as something that to due
(15:05):
to gingerly.
So it's gonna be veryinteresting to see.
And agreed.
I think Warren Buffet willprobably not be around long
enough to see that, but cuz juststatistically speaking but.
He does like banks and he doeslike insurance products.
And imagine Apple being hub forbanking and insurance products
and Warren Buffett sitting atthe helm.
(15:27):
I cannot think of a morepowerful platform to steward
retail products ever in thehistory of mankind.
Yes, agreed.
Completely agree.
Again and probably fairer, Ithink even for everybody around,
interestingly enough.
That's one would hope.
But I can, let me tell youwhat's happened in banking in a
small backward countries such asSpain, 20 years ago, there were,
(15:51):
Peter would probably know thenumbers better, but probably
about 80, 90 banks of national,of them with branches in Madrid.
Yeah.
Barcelona and some of thelargest cities.
And banking services werecompetitively priced because
there was a significant amountof competition.
(16:11):
Nowadays we're down to a handfulof banks that could, with large
market shares.
I think the top five banks inSpain probably have, I don't
know, 50, 60% market share.
Maybe probably know the numbersbetter.
I'd say more.
More so what's happened as aresult of the great financial
crisis and the European.
And Nike Crisis is that with anenormous consolidation.
(16:35):
And as a result of thatconsolidation, people expected
that regardless of where theECBs set the deposit rate they
would have Sorry.
They would, they would be ableto pay 0% for deposits.
This is the argument that nine,nine out of 10 bank analyst had
(16:57):
to buy European banks, that thejaws would work exceptionally
well in this cycle, that therewould be no increase in the
funding costs from deposits, andthen you would get all the
benefits of.
With loans and the reason theycould be sand about such a
strange prediction was thatthere's so much less
(17:19):
competition.
And they tested that in the UKmarket, which is a very
concentrated market.
And the, until the hiccups ofOctober and even after that,
there hasn't been a lot of thesetermination of bank deposits in
spite of everything.
I think that the svb and theother two banks with the plus
the pretty sweet situation mayhave changed that, sense of,
(17:43):
safety of having a bank depositfor many people.
And then this message fromApple, and if I'm right, and
Amazon comes out within the nextfew weeks and has its own
deposit product with some otherfinancial partner I, we are
going to have another financialcrisis in the making within a
couple of years I and focused ondeposits.
(18:05):
I think that's one scenario.
I think you're right that thereare fewer competitors, so less
competition.
Another element of what's goingon in Europe is the quantitative
quantitative easing, which isstill going on.
The money supply is the marketis still a wash in liquidity,
and it's not until June that theEuropean Central Bank is going
(18:27):
to actually go into quantitativetightening mode.
Let's see if that, let's see ifthat changes the dynamic on
deposits.
May I just say something onthat, which is that part of the
qualitative quantitative typingin Europe has.
It's too complicated for anormal person to understand,
which is the reduction in tlt ro Yeah.
(18:48):
Funding and that is going quitefast, right?
So Yeah.
And then it was one of the keyprograms for providing liquidity
to the banking system.
So maybe we can shift to a lessfinancial topic.
(19:08):
Let's look at ai.
Oh, fun.
What, Alex you wanna give us anupdate on how we're going and
getting to artificial generalintelligence?
Agi, so in a godlike manner, soenormously a rapid sub update.
I think everybody's seen,obviously stable diffusion six
months ago DLI and so forth.
(19:29):
On the image side, everybody acouple of months ago on the G P
T three, the three five, thenfour side.
And it's important to understandmore or less, what happened.
The most important thing tothink about, I think, and
understanding landscape is thatfor years people have been
publishing papers, working veryhard at trying to understand how
in a lab, how to use artificialintelligence.
(19:51):
And at the Genesis, this is back20 15, 20 16, OpenAI was built
to try to do this independentlyfrom the larger companies.
So again, Google, FacebookMicrosoft many people, one
after.
And Google has two dedicatedteams on ai.
Very powerful teams both goingafter it.
And the idea when Dali was firstreleased, which was a shot
(20:14):
across the bow to all the largercompanies.
Back in September and then whenG P T three was released and
quickly upgraded.
What happened was you nowstarted to have an open source,
a set of tools that people coulduse.
The pricing of the a p i of openAI was so low or could remain so
(20:34):
low that it provides for anenormous amount of capability
for people to go and try thingsand to basically offer.
AI ish or AI products, which arereally just the GPT APIs,
meaning that the ability to feedonto G P T, whatever it is that
you're doing.
So you've seen all kinds ofother companies larger companies
(20:55):
come out with tools, those thatwere working on AI before.
So things like Adobe andobviously Google and everybody
else has come up with stuff.
And then those that have justresold their bulk access or
wholesaling access to to openai.
The fact that it's open sourceis enormously powerful because
what it does is that it takesaway from people in lab coats
(21:17):
the control that they had insome very interesting interviews
of Eric Schmidt from a coupleyears ago, who was adamant that
all of his AI stuff, for all thereasons you know, that ethicist
will, will bring up, should bekept very controlled and inside
the lab and so on, so forth.
That cat is.
That Legion of cat is completelyout of bag by now.
(21:38):
So what you're seeing is thatyou're the main limita.
You're seeing a lot ofdevelopment in all kinds of
different ways.
Most of the activity on GitHub,which is the place where
developers particularly opensource of developers, share
code.
It's now all ai all the time, 24hours a day which is quite
interesting.
So you're starting to see a lotof things being developed.
The thing to think about rightnow, I think just to summarize
(22:00):
it and I'll open it up, is thereare three things I think there
are worth thinking about.
The first thing is that thesemodels have been trained a
certain particular with certainparticular set of data.
Some of it cooperated again witha cutoff date like about a year
ago.
I.
But that's not the future ofmodels.
Models are going to be nightlybuilds.
It's going to be something inwhich, Morgan Stanley or any
(22:22):
organization where they have aninternal one or an external one
facing customers are going tohave the, to retrain or to cont
continuously train on newinformation.
These master models that they'regonna have that're gonna make
available to customers or makeavailable to employees in order
to be able to find pretty muchanything that you have as
institutional knowledge in yourcompany.
And that's a gonna be a.
(22:42):
Fairly large business that'sgonna require a lot of effort
and handholding because again,you're democrat democratizing a
a technology.
The second thing is that rightnow there's this concept of how
many tokens, or let's say howmany words you can feed.
Into a thread.
So Che g p t, for example, tellsyou that they are threads and
people and that the threads willremember what you said.
(23:03):
That's not entirely true.
Basically, the way to thinkabout it is anywhere between, in
the beginning 2000 words, whichare really tokens, but let's
just say words for simplicity.
2000 words is its memory,meaning that if you give it.
10 times 200 words.
So if you ask it a question for200 words, then get a response
and ask it again, anotherquestion, or remember the first
200 words and actually theanswer they gave you.
(23:24):
So it has a little bit of memoryfor a short period of time, but
one of the biggest problems outthere is to try to give, its
more short.
Term or long-term memory.
And so you've seen thisdevelopment of what they call
vector databases, which arebasically ways in which you can
add some memory to the modelwith having to retrain it.
If you train the model, you addpermanently the data to the
(23:45):
model.
That's a good thing, but veryexpensive, very complicated.
But the idea is that you want tohave kind of a memory buffer.
The way to think about it islike if you have 10 years of
income statements or 50 productPDFs or anything like that you
really want those to beavailable, not, to, to j to your
AI to share G B T.
In order to be able to answercorrectly, again, because these
(24:06):
generalized models, can answergeneralized things, not specific
things about your knowledge.
And then the third piece, whichis fascinating is, which has
just started the last couple ofweeks, is this concept.
There's a very famous projectcalled Auto G P T, which has
been Having a lot of activityand recently, but basically is
the idea of user agents.
So the idea is that if youprompt a model meaning you write
(24:30):
it a job description saying youare an expert at marketing and
you know everything there is todo about the four Ps and you
know everything about all thisstuff, and you give it a
persona, then it's going toanswer in that particular way.
You can replicate that in abunch of different ways where
you spin up multiple models andyou give them different
personas.
So you can create yourself a c eO persona, which is, or a team
(24:54):
leader persona who is in whosejob it is to achieve certain
goals.
And then you have it interactwith other models who are
specialized in finance andoperations and marketing and HR
and whatever it is, or whateverthe components of your project
is.
And then you just, let it loose,give it a, a goal, and you just
(25:15):
watch, while they have all kindsof conversations about trying to
accomplish the goal at hand,again, it's very early.
But this gives you an insight asto how you can get really deep
insight and thoughtful insightfrom, from the systems, soon
enough.
So to close it out, this isenormously early.
It is impressive.
It has made people realize thisis not 10 years from, 10, 10
(25:37):
years away, or 20 years away.
And it is just as exciting as itis.
Scary.
That's all.
I'll say fascinating time.
Fascinating time.
What do you think about thepetition to Pause?
So I think it is a admirableview.
There's a couple of interestinginterviews.
Lex Frigman has a coupleinterviews, which are
(25:57):
interesting, if any, wasinterested on the reasons for
why you should take the time topause it.
It is laudable.
It is never gonna happen.
Okay.
I feel like I am the ludite, butI talked to Chad GBT often and I
have a subscription.
(26:17):
And so far from its own account,it tells me that, not in the
exact same words, but that it'sa very methodical librarian and
it doesn't have the ability todo any original thinking beyond
what it can find in the libraryand what I thought would happen.
(26:42):
And what is happening with chatGBTs.
Two different things I thoughtthat we were gonna go.
So this reminds me a little bitof what my friends were doing
out of engineering school in thelate eighties, early nineties,
that they were doing artificialintelligence as an ency effort
to a lot of information intodatabases so that a processor
(27:04):
could use the databases to comeout with answers that were stock
answers to stock question.
I thought when Anna and I wereroommates in New York and he
introduced me to a game calledSin City, that apparently he
developed some organizationalskills from having small now I
(27:26):
forget the words, but this is,goes back to the early work of
the Yes.
Mathematic.
Yep.
Which is that you have, fromsimilar initial conditions,
these, how do you call theseagents?
I forget.
Alex, these agents that wouldfind different paths to do
different things.
(27:47):
You can call'em agents.
Agents.
Agents, yeah.
And I thought that was far moreinteresting cause they would
come up with their own solutionby playing the game.
And when you get to.
The game of chess.
For instance, there were agentsthat came up with a way to play
chess that was systematicallyunbeatable.
Yep.
Some people said, okay chess isa, is a simple game because
(28:07):
there's always an optimalsolution to any position.
Let's try a different game.
And they try and go, which is aJapanese game of occupation, of
a territory from the enemy.
You might be committed with thatgame has white beads and black
beads and it is, they nobodythought that it would take, that
(28:27):
it would be easy to program acomputer to play Go.
And I think it was within like36 hours a program was able to
be the world.
Yeah.
And I thought for the, let meinterrupt you just for one
second.
Just to give you one piece ofinsight, which is helpful cuz I
remember we, we financed a bunchof companies that were trying to
(28:48):
do this kind of stuff about 20years ago, and it was a mess and
it was very much rule-based andso on, so forth.
But I would pause at the fall,the following thing to try to
reconcile the two views.
Your brain essentially learnsthe same way as a language
learning model.
Let's simplify, right?
So a deep learning model, andthe reason I say that is because
you as a child, experiencecertain things.
(29:09):
So for example, if you put yourhand.
Onto a hot stove, you willassociate or existing nerve
endings that were built earlier,even in your life about moving
your hand.
And then you will make aconnection with the fact that
touching the stove, obviouslycreates heat.
And I, if you were to be able tolive as many lives as.
(29:32):
A language learning model does,which is millions and millions
in lives in parallel, right?
So it's touching millions ofstoves and in certain particular
ways, in every possible way.
What's happening is, which isinteresting, is, and this is
your point, I think Compared tothe way people thought it would
be done, which is by, makingsure that we understand all the
rules, that we're a goodlibrarian, we know what section
(29:52):
of the library to go and lookfor whatever, where the
knowledge is stored.
It turns out that language thesemodels have are matrices of
numbers and weights.
That's all they are.
They're literally the equivalentof these connections that if
this happens there, then Ishould go this way, not that
way.
Kind of thing.
Super simple in some ways, butat a level that computing has
(30:16):
only been able to make availablein the last couple of years.
Let's just stuff that would'vebeen massively too big to do.
So the argument I would makejust in finishing is that we
have in some ways replicatedwith the way the human mind.
Develops at the early age andeven later stage in matrices in
(30:38):
a mathematical algorithm, trainthem.
And so I would argue that it isable to learn.
At least as well as a human can.
And I know that's a verypowerful statement.
We may not see it today but Ithink the argument is that,
(30:59):
that, we're, the way that itcomes up with the way it thinks
is very similar to the way thebrain, I think.
I think also if you, Luis,you're probably using it a lot
for our, the, for the types ofissues and topics that we get
into and are curious about, and.
And talk about and discuss.
If you ask it to write a shortpoem in the style of Elliot
(31:23):
about the restaurant across thestreet, I think you'd find you'd
get a, you'd get blown away byby what it can do stylistically.
And and, do I say it creatively?
Yeah.
Yeah.
Bloomberg trained the model andwhen you look at the paper, they
just did, released it last week.
And what's funny about it isthat the Bloomberg data, their
propriety data the Crown rules,those things are the most
(31:45):
important thing.
Only counter for 78 basis pointsfor less than 1% of the training
data they put into it.
So there's a wall to go beforeyou see something, but when that
model gets trained with actualBloomberg data, it'll be a site
to see.
I think Think that's, I thinkwe're gonna, we're gonna call it
(32:07):
a day and call it and say thanksto everyone.
Thanks very much.
Thanks, Luis.
And take care, Alex.
Thanks for organizing this.
Yes, thank you very much.
Thank you.
Bye.
Talk soon.
Bye.
Take care.
hello.
Good evening, good afternoon,good morning, wherever you are.
This is Off the Fence, atransatlantic podcast where
three guys in Madrid and in theNew York metropolitan area get
together and try to put theworld to this week we're gonna
talk about developments inbanking, in US banking and
specifically Apple offering asavings account.
(00:21):
It's gonna actually pay acompetitive interest rate which
should be interesting given theway deposits have been moving in
the US and and other and otherdevelopments.
But Luis, this was your topic.
Why didn't you, why don't youtake it, tell us what you think
what the issues are.
Thank you Peter.
I think, we discussed this topicin previous podcasts and.
(00:44):
My impression is that whenyou're presented with such a
fantastic spread betweendeposits that are site deposit
that pay nothing and federalconcentrated 5% or 500 basis
points spread between what bankspay for deposits and treasuries
short data treasuries, therewill be somebody somewhere that
(01:06):
will take advantage of that totake market share in deposits
and do something with it.
And we have, we, we talked aboutit and we couldn't, I couldn't
figure out who would be thedisruptor.
Apple yesterday broke the ice.
It is, in my opinion, a muchmore significant attacker than
any of the attackers that we sawin the early 2000, such as I n g
(01:28):
Direct or the online brokers.
Various reasons.
First of all, apple as an, asthe.
Owner of the operating system inwhich a lot of people's lives
work has an enormous amount ofsubscribers already many of whom
use different functions that theApple Pay app allows them to
(01:51):
use.
It is also a big contender inthe buy now pay later layaway
system of buying for retailpurchases.
And therefore the fact that theycan get funding at a competitive
rate for them and maybe not forthem, maybe somebody else who
does the actual financialmanagement of both the asset and
(02:12):
the liabilities of these in, ofthis book.
I think it's quite a threat tosome lenders to consu to some
consumer lenders in particular.
Unsecured consumer loans.
One has to believe that it won'tbe too long before Amazon thinks
that maybe this is also theirdomain, that they should
(02:35):
probably get into this.
And perhaps other of theplatforms that are involved in
either payments or consumerRetail might, will also develop
their own special, theinteresting savings products
with this.
I think the one belonged beforethis trend moves from west to
(02:57):
east as these trends generallydo.
And discuss with you guys inprevious podcast.
I believe that the currentexcitement with cyclical
Companies, specifically banksin, in the Eurozone might be a
little bit ahead of itselfbecause, this would be one of
the threats that we could thinkof for that markets.
(03:20):
What do you think?
I, go ahead Alex, please.
No, please, Peter.
I'm having a look on onlineright now.
What big banks.
Our paying is an average rate,and it's about 0.1%.
And most if you look acrosssmaller banks, they're, they get
to around 3%.
(03:41):
I see you have some specialoffers.
PNC has an account at 4%.
But this is a competitive move.
Banks are slow to raise depositrates.
We know that they're a lotquicker to raise their lending
rates.
It's which is natural.
Goldman Sachs, who will beservicing this, took a stab at
(04:02):
retail banking in a, with a unitcalled Marcus that hasn't gone
so well.
It hasn't been unwound, but it'sbeen it's been reduced that
back.
Apple and Goldman have beenworking together for a while.
And especially on the card side.
I, and, but this is a new andinteresting move.
It's interesting also causeApple has none of the capital or
(04:25):
I believe liquidity concernsthat other banks might have in,
in backing up a savings account.
And this has become an issue inthe last few, we, few months as
because Yeah.
What sovereign Silicon ValleyBank was was liquidity.
The liquidity can moveincredibly rapidly.
And this is a huge concern forregulators and supervisors.
(04:47):
Supervisors right now, I don'tthink, I don't think Apple
offers a deposit cause Appledoesn't own a bank.
I think the deposit have to lookinto the small print is probably
deposit at Goldman Sachs or somethird party bank.
Apple just is the facilitator.
I think.
The marketer.
Yeah.
But that's the whole point,right?
So few things.
That's right.
So the re what I'm trying tosay, Alex, is that the
(05:09):
regulatory constraints on thedeposits will be with the
department taking bank will bewith Goldman.
That, that's the whole question.
The funny part is that Marcusoffers today 3.9% versus four
and change percent.
You're not gonna your ownproduct are you?
Yeah, of course.
Exactly.
You're gonna love that.
And it shows you the power ofthe thing.
So a few things you mentioned atthe beginning, the impetus for
(05:32):
this and it makes sense is, thefact that the spreads have widen
significantly and that the banksbeing slow to raise rates, as
Peter mentioned.
Given that we've had, a good 10years of customers being used to
not earning anything in theirdeposit accounts.
This is, marketing wise kind ofan important move for a.
National player as opposed to aregional player.
(05:52):
Regional recently certainly havehad to fight for the deposits by
paying higher rates.
It's a move that makes a wholelot of sense.
And then actually like similarto SVB in some ways, what you
worry about is with an inflow ofsignificant amount of savings in
all one swoop in a fairly bigmarketing push by a very
important player that has a lotof customers, you really wonder
(06:13):
where, where those deposits aregoing to go and who's taking the
gap book decision.
You'd hope that it's not Apple,as you mentioned, since they
don't have expertise,credibility, or even a structure
to do it.
So as you say, most likelysomebody else is doing it.
And if it's Goldman, fantastic.
It's Goldman Sachs, I mean atthe Apple.
There you go.
(06:33):
Starting today, apple Card userscan choose to grow their daily
cash rewards with savingsaccount from Goldman Sachs,
which offers high yield annualpercent in yield, annual
percentage yield of four point15% rate.
That's one than 10 times thenational average.
No fees, no minimum deposits.
Yeah.
It's a perfect marketing thing,but the argument that I'd like
to make, just to move theconversation elsewhere is that
(06:54):
for sure we have an opportunityright now because we've had
this, the rates back up andexpectations have been very low
and it takes a while forcustomers.
Customers will look at this asbeing a good deal.
That's great, but my argument isa different one, which is that
Apple should.
Itself, long term B, in thetransaction processing business,
because one of the main factorsthat keeps the wholesale rate of
(07:16):
transaction processing in the USat 2% right is fraud.
And if there's one thing thatthe marriage of software and
hardware in your hand, has beenable to handle is biometric
verification and really gettingan end-to-end between the
receiver and the sender of a ifnot trusted, but.
Verified identity ecosystem,which is quite strong.
(07:41):
The apple has both the physicalhardware and the software to run
transactions.
Again, forget crypto, forget,how it gets handled.
But the idea is for them to be apayment processor, given that
they can ensure what is probablythe largest piece.
Of the puzzle in terms of thecost of wholesaling, transaction
processing I think long termease and attractive is an
(08:03):
attractive business.
It could very well be that theywanna grow their marketing shops
at this point, which makessense, and turn that into over
time, building a real financialbusiness where they do
transaction process.
Not take, necessarily back andbalance sheet risk, but just be
there to make sure that you havethe.
Funds in your account and thatyou're making an accurate
payment.
That to me, I think is a longterm play That would be a
(08:25):
significant source of growth forApple.
And what does it mean for banks?
I don't know if you can hear me.
Can you hear me?
We can.
Yes.
And just because of that, I'mgonna answer it.
The the problem with banks, Ithink is that the physical
existence of a bank branch mayhave outlived its relevance.
(08:46):
And as such, you have a lot oflegacy costs associated in a
cost structure that makes itfairly difficult to, to compete.
Those kind of banks are gonnabe, this is not a new theme.
People have been talking aboutit for 20 years and it hasn't
happened, of course.
But I think more and more mysense is particularly as you get
to alternative ways ofprocessing transaction, which
are almost costless, it's gonnabe very difficult for people to
(09:10):
accept the fees associated withtraditional banking.
That's, and that I agreecompletely.
And the, and place where feesare most obvious is in payments.
Yep.
And for years, banks basicallyhad a monopoly on payments.
They no longer do.
Yet there's still competitive,there's still room for new
competition in payments and andApple is part of it.
(09:34):
Others are in the field and andbanks are feeling it and bank,
especially in Europe.
Agreed, yep.
I think if I may say somethingvery obvious the big difference
between.
A bank like i n g direct that inthe early nineties started
collecting deposits by paying acompetitive rate and then didn't
(09:56):
know what to do with thosedeposits.
And the current situation isthat Apple also originates an
asset product, which is the,these layaway loans.
And they are the platform whereyou can do both.
Whoever they partner with thatruns the both portfolios.
We'll be very keen to see that,this works as, as well as it
(10:19):
does into the advantage of Appleand the financial partner they
have.
And I think that if theycontinue on this vein and their
opportunity for other products,it's probably different in
various parts of the world, but,I am going to go to meet some
people from Apple because, thereis an enormous opportunity for a
(10:40):
brand like Apple to develop amass affluent independent
financial advisor business inEurope, which is probably one of
the areas of the market wherethere's the most fat in the
world.
Yes.
And and this would be terrificfor the vast majority of the.
Population of the European Unionor and other Europeans.
But what I would also remind youof is that a few years ago,
(11:04):
Michael Milken, in one of hisappearances at one of the
conferences with, asked toprovide advice to some of these
platforms.
And what he told them is, don'tget into finance because the
regulation will bug you down.
I talked this morning I shouldcorrect myself.
I exchanged WhatsApps thismorning with one of the most
(11:27):
senior people that I have accessto WhatsApp at eight o'clock in
the morning in finance.
He was the chairman of a bank inSpain.
He was the vice chairman of oneof the top four investment banks
for Europe.
And I asked him what, where hethoughts about this, and one of
the things he said is, thiscompany was doing very well, UN,
(11:47):
until or has done very well.
Let's hope they don't want toget into the MOAs of being
regulated as a financial entity.
And I, and my, what I understandis great about this deal with
Goldman Sachs is that GoldmanSachs already pays all the fixed
costs of being regulated as afinancial entity and if they can
get the additional business, Ofbeing Apple's joint venture
(12:11):
partner in this, everybody'sbetter off and including the
public probably.
So I think it's a, it's anincredible, incredibly I think
it is as material to the bankingto the retail banking world as
the cash management account in1973 was to the retail banking
(12:32):
world when Marin h introducedthat.
I think you're right.
And there's something else aboutit, which is in recent years I
have heard an argument inbanking, which goes something
like this.
A bank does three things.
It takes deposits, it lendsmoney, and it makes payments.
It affects payments.
The payments business it's goingit's, if it's not gone, it's
(12:55):
going.
Others have entered.
You don't really need a bankinglicense to do it.
They're companies that can do itwithout the legacy.
The lending business is also isalso going private equity and
others are now lend money.
Amazon lends money to small andmedium enterprises and
securitize to to fund it.
(13:16):
There's a lot going on in thelending space.
Buy now, pay later is done by anon-bank, bank financial
institution or can be done.
So there's a lot going on inthat space.
But deposits, taking deposits iswhere the regulation kicks in
when you're a deposit takinginstitution, and thus others
weren't gonna go into it.
(13:36):
The, they may try to createAnna, the instruments that look
like deposits, feel likedeposits, but they're not gonna
be able to offer customersguaranteed deposits because they
don't wanna be regulated.
As Luis has just pointed out.
This seems to be a way in.
You, this partnership withGoldman Sachs to to get a big
tech into that space, into thedeposit taking space.
(13:59):
And these will be depositsbecause I'm, they'll be
guaranteed by the F D I C.
So I, I agree.
It's, it this is potentially abig move.
I agree.
I think what's interesting abouthere is you have two best of
breeds, honestly, in, in mymind, between Apple being
knowledgeable on the softwarehardware side, consumer loyalty,
(14:20):
privacy brand and obviouslyGoldman Sachs on probably
regulatory management and alldifferent financial
capabilities.
So you have two best of breeds,which for the moment early are,
have compatible.
And corresponding strengths intrying to address this.
My, my question would be longterm, I think to both your
(14:43):
points, how forward integratedinto this, does Zapp want to be
long term?
Right now it's easy.
It's a marketing deal.
That's no problem.
Longer term, it's gonna be abusiness for them.
But if it is, it's a completelydifferent business than what
they do and would be, enormouslyrisky.
And I think for the investmentbase, for Warren Buffet, for
everybody else around would beviewed as something that to due
(15:05):
to gingerly.
So it's gonna be veryinteresting to see.
And agreed.
I think Warren Buffet willprobably not be around long
enough to see that, but cuz juststatistically speaking but.
He does like banks and he doeslike insurance products.
And imagine Apple being hub forbanking and insurance products
and Warren Buffett sitting atthe helm.
(15:27):
I cannot think of a morepowerful platform to steward
retail products ever in thehistory of mankind.
Yes, agreed.
Completely agree.
Again and probably fairer, Ithink even for everybody around,
interestingly enough.
That's one would hope.
But I can, let me tell youwhat's happened in banking in a
small backward countries such asSpain, 20 years ago, there were,
(15:51):
Peter would probably know thenumbers better, but probably
about 80, 90 banks of national,of them with branches in Madrid.
Yeah.
Barcelona and some of thelargest cities.
And banking services werecompetitively priced because
there was a significant amountof competition.
(16:11):
Nowadays we're down to a handfulof banks that could, with large
market shares.
I think the top five banks inSpain probably have, I don't
know, 50, 60% market share.
Maybe probably know the numbersbetter.
I'd say more.
More so what's happened as aresult of the great financial
crisis and the European.
And Nike Crisis is that with anenormous consolidation.
(16:35):
And as a result of thatconsolidation, people expected
that regardless of where theECBs set the deposit rate they
would have Sorry.
They would, they would be ableto pay 0% for deposits.
This is the argument that nine,nine out of 10 bank analyst had
(16:57):
to buy European banks, that thejaws would work exceptionally
well in this cycle, that therewould be no increase in the
funding costs from deposits, andthen you would get all the
benefits of.
With loans and the reason theycould be sand about such a
strange prediction was thatthere's so much less
(17:19):
competition.
And they tested that in the UKmarket, which is a very
concentrated market.
And the, until the hiccups ofOctober and even after that,
there hasn't been a lot of thesetermination of bank deposits in
spite of everything.
I think that the svb and theother two banks with the plus
the pretty sweet situation mayhave changed that, sense of,
(17:43):
safety of having a bank depositfor many people.
And then this message fromApple, and if I'm right, and
Amazon comes out within the nextfew weeks and has its own
deposit product with some otherfinancial partner I, we are
going to have another financialcrisis in the making within a
couple of years I and focused ondeposits.
(18:05):
I think that's one scenario.
I think you're right that thereare fewer competitors, so less
competition.
Another element of what's goingon in Europe is the quantitative
quantitative easing, which isstill going on.
The money supply is the marketis still a wash in liquidity,
and it's not until June that theEuropean Central Bank is going
(18:27):
to actually go into quantitativetightening mode.
Let's see if that, let's see ifthat changes the dynamic on
deposits.
May I just say something onthat, which is that part of the
qualitative quantitative typingin Europe has.
It's too complicated for anormal person to understand,
which is the reduction in tlt ro Yeah.
(18:48):
Funding and that is going quitefast, right?
So Yeah.
And then it was one of the keyprograms for providing liquidity
to the banking system.
So maybe we can shift to a lessfinancial topic.
(19:08):
Let's look at ai.
Oh, fun.
What, Alex you wanna give us anupdate on how we're going and
getting to artificial generalintelligence?
Agi, so in a godlike manner, soenormously a rapid sub update.
I think everybody's seen,obviously stable diffusion six
months ago DLI and so forth.
(19:29):
On the image side, everybody acouple of months ago on the G P
T three, the three five, thenfour side.
And it's important to understandmore or less, what happened.
The most important thing tothink about, I think, and
understanding landscape is thatfor years people have been
publishing papers, working veryhard at trying to understand how
in a lab, how to use artificialintelligence.
(19:51):
And at the Genesis, this is back20 15, 20 16, OpenAI was built
to try to do this independentlyfrom the larger companies.
So again, Google, FacebookMicrosoft many people, one
after.
And Google has two dedicatedteams on ai.
Very powerful teams both goingafter it.
And the idea when Dali was firstreleased, which was a shot
(20:14):
across the bow to all the largercompanies.
Back in September and then whenG P T three was released and
quickly upgraded.
What happened was you nowstarted to have an open source,
a set of tools that people coulduse.
The pricing of the a p i of openAI was so low or could remain so
(20:34):
low that it provides for anenormous amount of capability
for people to go and try thingsand to basically offer.
AI ish or AI products, which arereally just the GPT APIs,
meaning that the ability to feedonto G P T, whatever it is that
you're doing.
So you've seen all kinds ofother companies larger companies
(20:55):
come out with tools, those thatwere working on AI before.
So things like Adobe andobviously Google and everybody
else has come up with stuff.
And then those that have justresold their bulk access or
wholesaling access to to openai.
The fact that it's open sourceis enormously powerful because
what it does is that it takesaway from people in lab coats
(21:17):
the control that they had insome very interesting interviews
of Eric Schmidt from a coupleyears ago, who was adamant that
all of his AI stuff, for all thereasons you know, that ethicist
will, will bring up, should bekept very controlled and inside
the lab and so on, so forth.
That cat is.
That Legion of cat is completelyout of bag by now.
(21:38):
So what you're seeing is thatyou're the main limita.
You're seeing a lot ofdevelopment in all kinds of
different ways.
Most of the activity on GitHub,which is the place where
developers particularly opensource of developers, share
code.
It's now all ai all the time, 24hours a day which is quite
interesting.
So you're starting to see a lotof things being developed.
The thing to think about rightnow, I think just to summarize
(22:00):
it and I'll open it up, is thereare three things I think there
are worth thinking about.
The first thing is that thesemodels have been trained a
certain particular with certainparticular set of data.
Some of it cooperated again witha cutoff date like about a year
ago.
I.
But that's not the future ofmodels.
Models are going to be nightlybuilds.
It's going to be something inwhich, Morgan Stanley or any
(22:22):
organization where they have aninternal one or an external one
facing customers are going tohave the, to retrain or to cont
continuously train on newinformation.
These master models that they'regonna have that're gonna make
available to customers or makeavailable to employees in order
to be able to find pretty muchanything that you have as
institutional knowledge in yourcompany.
And that's a gonna be a.
(22:42):
Fairly large business that'sgonna require a lot of effort
and handholding because again,you're democrat democratizing a
a technology.
The second thing is that rightnow there's this concept of how
many tokens, or let's say howmany words you can feed.
Into a thread.
So Che g p t, for example, tellsyou that they are threads and
people and that the threads willremember what you said.
(23:03):
That's not entirely true.
Basically, the way to thinkabout it is anywhere between, in
the beginning 2000 words, whichare really tokens, but let's
just say words for simplicity.
2000 words is its memory,meaning that if you give it.
10 times 200 words.
So if you ask it a question for200 words, then get a response
and ask it again, anotherquestion, or remember the first
200 words and actually theanswer they gave you.
(23:24):
So it has a little bit of memoryfor a short period of time, but
one of the biggest problems outthere is to try to give, its
more short.
Term or long-term memory.
And so you've seen thisdevelopment of what they call
vector databases, which arebasically ways in which you can
add some memory to the modelwith having to retrain it.
If you train the model, you addpermanently the data to the
(23:45):
model.
That's a good thing, but veryexpensive, very complicated.
But the idea is that you want tohave kind of a memory buffer.
The way to think about it islike if you have 10 years of
income statements or 50 productPDFs or anything like that you
really want those to beavailable, not, to, to j to your
AI to share G B T.
In order to be able to answercorrectly, again, because these
(24:06):
generalized models, can answergeneralized things, not specific
things about your knowledge.
And then the third piece, whichis fascinating is, which has
just started the last couple ofweeks, is this concept.
There's a very famous projectcalled Auto G P T, which has
been Having a lot of activityand recently, but basically is
the idea of user agents.
So the idea is that if youprompt a model meaning you write
(24:30):
it a job description saying youare an expert at marketing and
you know everything there is todo about the four Ps and you
know everything about all thisstuff, and you give it a
persona, then it's going toanswer in that particular way.
You can replicate that in abunch of different ways where
you spin up multiple models andyou give them different
personas.
So you can create yourself a c eO persona, which is, or a team
(24:54):
leader persona who is in whosejob it is to achieve certain
goals.
And then you have it interactwith other models who are
specialized in finance andoperations and marketing and HR
and whatever it is, or whateverthe components of your project
is.
And then you just, let it loose,give it a, a goal, and you just
(25:15):
watch, while they have all kindsof conversations about trying to
accomplish the goal at hand,again, it's very early.
But this gives you an insight asto how you can get really deep
insight and thoughtful insightfrom, from the systems, soon
enough.
So to close it out, this isenormously early.
It is impressive.
It has made people realize thisis not 10 years from, 10, 10
(25:37):
years away, or 20 years away.
And it is just as exciting as itis.
Scary.
That's all.
I'll say fascinating time.
Fascinating time.
What do you think about thepetition to Pause?
So I think it is a admirableview.
There's a couple of interestinginterviews.
Lex Frigman has a coupleinterviews, which are
(25:57):
interesting, if any, wasinterested on the reasons for
why you should take the time topause it.
It is laudable.
It is never gonna happen.
Okay.
I feel like I am the ludite, butI talked to Chad GBT often and I
have a subscription.
(26:17):
And so far from its own account,it tells me that, not in the
exact same words, but that it'sa very methodical librarian and
it doesn't have the ability todo any original thinking beyond
what it can find in the libraryand what I thought would happen.
(26:42):
And what is happening with chatGBTs.
Two different things I thoughtthat we were gonna go.
So this reminds me a little bitof what my friends were doing
out of engineering school in thelate eighties, early nineties,
that they were doing artificialintelligence as an ency effort
to a lot of information intodatabases so that a processor
(27:04):
could use the databases to comeout with answers that were stock
answers to stock question.
I thought when Anna and I wereroommates in New York and he
introduced me to a game calledSin City, that apparently he
developed some organizationalskills from having small now I
(27:26):
forget the words, but this is,goes back to the early work of
the Yes.
Mathematic.
Yep.
Which is that you have, fromsimilar initial conditions,
these, how do you call theseagents?
I forget.
Alex, these agents that wouldfind different paths to do
different things.
(27:47):
You can call'em agents.
Agents.
Agents, yeah.
And I thought that was far moreinteresting cause they would
come up with their own solutionby playing the game.
And when you get to.
The game of chess.
For instance, there were agentsthat came up with a way to play
chess that was systematicallyunbeatable.
Yep.
Some people said, okay chess isa, is a simple game because
(28:07):
there's always an optimalsolution to any position.
Let's try a different game.
And they try and go, which is aJapanese game of occupation, of
a territory from the enemy.
You might be committed with thatgame has white beads and black
beads and it is, they nobodythought that it would take, that
(28:27):
it would be easy to program acomputer to play Go.
And I think it was within like36 hours a program was able to
be the world.
Yeah.
And I thought for the, let meinterrupt you just for one
second.
Just to give you one piece ofinsight, which is helpful cuz I
remember we, we financed a bunchof companies that were trying to
(28:48):
do this kind of stuff about 20years ago, and it was a mess and
it was very much rule-based andso on, so forth.
But I would pause at the fall,the following thing to try to
reconcile the two views.
Your brain essentially learnsthe same way as a language
learning model.
Let's simplify, right?
So a deep learning model, andthe reason I say that is because
you as a child, experiencecertain things.
(29:09):
So for example, if you put yourhand.
Onto a hot stove, you willassociate or existing nerve
endings that were built earlier,even in your life about moving
your hand.
And then you will make aconnection with the fact that
touching the stove, obviouslycreates heat.
And I, if you were to be able tolive as many lives as.
(29:32):
A language learning model does,which is millions and millions
in lives in parallel, right?
So it's touching millions ofstoves and in certain particular
ways, in every possible way.
What's happening is, which isinteresting, is, and this is
your point, I think Compared tothe way people thought it would
be done, which is by, makingsure that we understand all the
rules, that we're a goodlibrarian, we know what section
(29:52):
of the library to go and lookfor whatever, where the
knowledge is stored.
It turns out that language thesemodels have are matrices of
numbers and weights.
That's all they are.
They're literally the equivalentof these connections that if
this happens there, then Ishould go this way, not that
way.
Kind of thing.
Super simple in some ways, butat a level that computing has
(30:16):
only been able to make availablein the last couple of years.
Let's just stuff that would'vebeen massively too big to do.
So the argument I would makejust in finishing is that we
have in some ways replicatedwith the way the human mind.
Develops at the early age andeven later stage in matrices in
(30:38):
a mathematical algorithm, trainthem.
And so I would argue that it isable to learn.
At least as well as a human can.
And I know that's a verypowerful statement.
We may not see it today but Ithink the argument is that,
(30:59):
that, we're, the way that itcomes up with the way it thinks
is very similar to the way thebrain, I think.
I think also if you, Luis,you're probably using it a lot
for our, the, for the types ofissues and topics that we get
into and are curious about, and.
And talk about and discuss.
If you ask it to write a shortpoem in the style of Elliot
(31:23):
about the restaurant across thestreet, I think you'd find you'd
get a, you'd get blown away byby what it can do stylistically.
And and, do I say it creatively?
Yeah.
Yeah.
Bloomberg trained the model andwhen you look at the paper, they
just did, released it last week.
And what's funny about it isthat the Bloomberg data, their
propriety data the Crown rules,those things are the most
(31:45):
important thing.
Only counter for 78 basis pointsfor less than 1% of the training
data they put into it.
So there's a wall to go beforeyou see something, but when that
model gets trained with actualBloomberg data, it'll be a site
to see.
I think Think that's, I thinkwe're gonna, we're gonna call it
(32:07):
a day and call it and say thanksto everyone.
Thanks very much.
Thanks, Luis.
And take care, Alex.
Thanks for organizing this.
Yes, thank you very much.
Thank you.
Bye.
Talk soon.
Bye.
Take care.
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy And Charlamagne Tha God!
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.