The Real AI: Unmasking the Hype | Kevin Surace

[00:00:06] There is so much hype about AI now.

[00:00:10] What really bothers me is when I see all these people giving quote-unquote master classes in AI

[00:00:15] when they have no experience in AI, no experience in technology.

[00:00:19] They've never mentioned the word AI before in their lives.

[00:00:22] Then there's another group of people saying AI is going to destroy the world.

[00:00:25] It's going to take jobs. It's going to do this. It's going to do that.

[00:00:27] I've been working in AI. I was an undergrad. I worked on AI.

[00:00:30] I went to grad school for AI. I had AI related jobs for many years,

[00:00:34] including in the investing business.

[00:00:36] But you've all heard from me.

[00:00:38] I wanted to bring on a friend of mine who is a pioneer in the space of AI.

[00:00:42] You might have heard of a speech recognition product called Siri.

[00:00:46] He helped develop that. He helped develop many other AI projects.

[00:00:49] He has a set of AI companies right now.

[00:00:52] Here's Kevin Surace. We talk a little bit about the death of cities.

[00:00:56] People always want to talk to me about that, and I understand because I've written about that.

[00:01:00] Then we get into the heart of the matter.

[00:01:04] What is the truth and what is the hype on AI

[00:01:07] and how can you use it to make your life better?

[00:01:10] This isn't your average business podcast.

[00:01:16] And he's not your average host.

[00:01:18] This is The James Altiger Show.

[00:01:30] Yeah, we're going to talk about AI, but Kevin and I started talking about how cities are doing.

[00:01:35] I don't think it's necessarily bad for the country when cities are damaged permanently,

[00:01:41] like New York City, San Francisco, Chicago, and so on,

[00:01:43] because the talent disperses throughout the rest of the country.

[00:01:46] Right. That is indeed what's happening.

[00:01:49] It is working out, but if I were in commercial real estate and hanging on to something at a certain value,

[00:01:55] probably that's not the value going forward.

[00:01:57] That value may be cut in half.

[00:01:59] No, look at San Francisco. The big real estate developers and private...

[00:02:03] They're just handing over the keys to the bank.

[00:02:06] They're saying, look, we're good guys.

[00:02:08] You take over the building. Here are the keys.

[00:02:10] Did you see that with the big mall in San Francisco where everybody walked out basically and said,

[00:02:16] it's a mall developer. It's Simon malls.

[00:02:18] And I think it's Simon, isn't it? And they handed over the keys.

[00:02:21] Yeah.

[00:02:22] Handed over the keys to the bank.

[00:02:23] They said, we don't want anything to do with it. You can just have it. That's kind of crazy.

[00:02:27] I know. It's astonishing.

[00:02:28] Like they pour hundreds of millions of dollars into this mall and they're like,

[00:02:31] eh, didn't work out.

[00:02:33] It costs us more to keep going than to just shut it down.

[00:02:37] So it's amazing.

[00:02:38] What's the bank going to do? Like the bank can't sell it if this mall owner got,

[00:02:42] who's going to buy it?

[00:02:43] You know what? Banks don't like to be in the real estate business as we learned back

[00:02:47] when we had the real estate crash, right?

[00:02:49] And everything falls on banks.

[00:02:50] The first thing they want to do is get rid of it at any cost.

[00:02:52] They do not want to carry that on their books.

[00:02:54] Yeah. So maybe there's a bargain, but even if the bargain hunter buys it,

[00:02:57] what does he do? Turn it into an ice skating rink?

[00:02:59] Like it's not going to make any money.

[00:03:01] San Francisco has a retail problem because people don't want to go to the shop

[00:03:05] because of the homeless situation and a right of other things, right?

[00:03:08] It's during the day.

[00:03:10] The offices are half populated.

[00:03:12] So that cuts that in half.

[00:03:14] And at night, nobody wants to go there on the weekends.

[00:03:16] They only go there. It's a real problem, right?

[00:03:18] Yeah. I mean, look, it's like New York City though.

[00:03:21] Like the good thing about San Francisco is the residential areas kind of,

[00:03:24] you know, around the downtown are fine.

[00:03:27] Just like New York City, you go to the outlying boroughs.

[00:03:29] It's fine.

[00:03:30] But the bulk of the money, like let's say Midtown and Wall Street

[00:03:34] in New York City is dead.

[00:03:36] So how does this long-term effect the future of the city?

[00:03:39] I think we don't really know, but there's going to be a significant problems.

[00:03:43] It's a significant problem.

[00:03:45] So you moved to Florida in the pandemic.

[00:03:47] I was in Florida for a while now. I'm outside of Atlanta, Georgia.

[00:03:49] What about you? Are you still in San Francisco?

[00:03:52] Right now I'm in upstate New York.

[00:03:54] We still have a place in California as well,

[00:03:56] but summer and fall, this is a stunning place to be.

[00:03:59] Why not go to a place warmer like, you know, Florida?

[00:04:03] Because we have a place in Maui for the winter, so I don't need Florida.

[00:04:06] Oh, okay. Well, there you go.

[00:04:08] With all due respect to Florida and its politics, you know, no, Maui is pretty good.

[00:04:13] All right. Good for you.

[00:04:14] So look, Kevin, you and I, we've spoken about AI years ago, long before the current,

[00:04:20] I don't want to say it's hype because it's very real.

[00:04:22] The innovations in large language models.

[00:04:24] But what's hype?

[00:04:26] We haven't spoken since about AI.

[00:04:28] What's hype?

[00:04:29] And I'm curious if you agree.

[00:04:30] I can't stand it.

[00:04:31] All these people who are the last they were doing was marketing,

[00:04:35] whatever it is that they're marketing.

[00:04:36] Now they're like, hey, join our masterclass mastermind, you know,

[00:04:41] six day retreat to learn how to make a million dollars in a week on using AI.

[00:04:46] Like, you know, how I made seven figures in three days using AI.

[00:04:51] I can't stand it.

[00:04:52] Like I've been working on AI related stuff since literally 1987.

[00:04:59] And I know you're probably even longer than that.

[00:05:02] Since the 90s.

[00:05:03] Yeah.

[00:05:04] Since the 90s.

[00:05:05] Yeah, that's right.

[00:05:06] So and I'm an investor in your AI startup, so what's going on?

[00:05:11] That's right.

[00:05:12] What's your take?

[00:05:13] So look, here's my take.

[00:05:14] First of all for the general public out there, right?

[00:05:16] We've been working on an AI since the 50s as you know.

[00:05:19] This is not new.

[00:05:20] There's been nuclear winters of AI death.

[00:05:24] But the fact of the matter is every step towards large language models

[00:05:28] happened because of prior steps, right?

[00:05:30] Every single one.

[00:05:31] Like we understood neural nets decades ago, but we couldn't do deep neural nets

[00:05:36] until about 11 or 12 years ago because we just didn't have the compute horsepower

[00:05:40] and the cloud power and everything else.

[00:05:42] Just to put that in perspective, even the large language model of open AI,

[00:05:46] chat GPT as we know it now, it took a year and a half of basically

[00:05:50] super computers crunching the data with neural networks

[00:05:53] before we had a first draft and then another year and a half of supervised

[00:05:57] learning with humans going back and forth with it.

[00:06:01] So it was a year and a half of computational power with the latest

[00:06:04] super computers that got done and what we needed to get done,

[00:06:07] which is longer than it takes for a computer vision model and so on.

[00:06:10] Absolutely.

[00:06:11] I mean only someone who had a billion dollars of Azure time

[00:06:16] and a billion dollar investment from Microsoft could have pulled that off.

[00:06:20] So it's not like some startup could just go and say,

[00:06:23] hey I'm going to build a large language model that has a trillion tokens.

[00:06:27] It's not going to happen, right?

[00:06:28] And so they had a unique situation where Microsoft said,

[00:06:31] we will give you free use of Azure.

[00:06:33] You can use anything you want and I'll give you a billion dollars.

[00:06:36] Go for it.

[00:06:37] And that allowed them in a very unusual startup way to say,

[00:06:41] I'll just spend all the money and hopefully when you have a trillion

[00:06:47] tokens you're going to get something interesting.

[00:06:49] Now that's a huge neural net as you know and then it turns out

[00:06:52] you get this huge neural net and it spews out a fair bit of garbage

[00:06:56] because of course it was unsupervised.

[00:06:58] Unsupervised doesn't mean a human was watching it,

[00:07:01] as you know I'm telling this for your listeners.

[00:07:03] Unsupervised means that all the information isn't tagged

[00:07:06] say whether it's real or fiction or whatever.

[00:07:08] So it's reading fiction, it's reading fact,

[00:07:10] it's reading garbage too, it reads everything, right?

[00:07:12] And of course it doesn't know necessarily that what it just read

[00:07:16] was fiction, a novel.

[00:07:18] And so at first if you don't put guardrails around it

[00:07:21] and you say, do you love me?

[00:07:24] Of course it's going to build a sentence that says,

[00:07:26] of course I love you, I'm madly in love with you.

[00:07:28] And you go, it's sentient, no it happened to read

[00:07:31] 100,000 fiction novels.

[00:07:33] Of course it knows how to say I love you, right?

[00:07:35] It has no sentience whatsoever.

[00:07:37] And so then as you know open AI for more than a year

[00:07:40] hired 1,000 people in Turkey, if I remember.

[00:07:44] And to start building a rules engine around it.

[00:07:47] Now what's funny is before we had deep learning at all

[00:07:51] we built the entire AI infrastructure of a variety of

[00:07:54] rules engines, that's all we had, right?

[00:07:56] So we didn't have the learning so we just said

[00:07:58] I'll just build rules engines.

[00:08:00] So when I was building the very first kind of versions

[00:08:03] of Siri for general magic in the 90s and that's

[00:08:06] what Siri was based on is all that work that we did.

[00:08:09] It was hidden Markov models for speech recognition

[00:08:12] and then it was all these big huge rule models

[00:08:15] basically that said if you asked, her name was Mary,

[00:08:18] if you asked Mary do you love me,

[00:08:21] we would have all these randomized responses

[00:08:23] that were put together but there was a rules engine

[00:08:25] that was doing that.

[00:08:26] And of course people heard that and they said it's AI.

[00:08:28] Well that's the computer horsepower we had.

[00:08:31] So and Kevin just on that, I didn't know you worked

[00:08:33] on at General Magic on Siri because I worked

[00:08:36] for a while with Kai Fuli who used, you know,

[00:08:39] developed the whole hidden Markov models to recognize

[00:08:43] what your voice was saying.

[00:08:45] So it would use this very sophisticated statistics

[00:08:47] to recognize his basic project was funded

[00:08:50] by the Department of Navy.

[00:08:52] So the first example of hidden Markov models

[00:08:55] recognizing speech was fire the missiles

[00:08:58] and things like that clean the deck.

[00:09:01] And then he was over at Apple while you were

[00:09:04] at General Magic presumably and that's right.

[00:09:07] Just a point of trivia.

[00:09:09] I was in Norway a few weeks ago and I met this former

[00:09:12] weather girl, weather woman named Siri who apparently

[00:09:16] was, Siri was named after her because the guy

[00:09:19] I guess at General Magic or Apple or whatever

[00:09:21] who was the product leader, hypothetically,

[00:09:24] I will say this is my words had a crush on her

[00:09:27] and called her and potentially, you know,

[00:09:30] she said let him answer this question for sure

[00:09:33] but potentially named Siri after this woman.

[00:09:35] Before Siri's there was Mary and Mary is named after

[00:09:38] Mary McDonald and Mary was our voice artist for Mary.

[00:09:43] And that was, by the way, that became GM on Star.

[00:09:46] That's what we built right?

[00:09:48] And so we had millions and millions of users on this system

[00:09:50] that were all using speech recognition and it

[00:09:53] and she could read your email and get your calendar

[00:09:55] and answer your phone for you which is really cool.

[00:09:57] She'd answer the phone and say, oh hi Kevin

[00:10:00] I recognize the number what would you like to do?

[00:10:02] I'd like to get on so and so's calendar.

[00:10:04] No problem put you on James' calendar.

[00:10:06] When are you available?

[00:10:07] Like all of that existed in the 90s, right?

[00:10:09] But even with those hit Markov models what we would do

[00:10:13] is we hired, we were the first to ever do this

[00:10:15] a whole bunch of linguists that would literally listen

[00:10:18] to what you said and codify other ways to recognize it.

[00:10:23] So I'm going to give you an example.

[00:10:25] We recognized that if you said read my email

[00:10:28] we knew what to do, read your email,

[00:10:30] and then we would get to the first one.

[00:10:32] But some people would then say, get me my email.

[00:10:34] I want my mail. Where is my mail?

[00:10:36] And you go there's a lot of ways to ask for getting your email.

[00:10:39] And so you had to codify those through linguists

[00:10:42] and in fact when then Ciri started

[00:10:44] which was based on all the work we did at General Magic

[00:10:47] and then that got sold to Apple.

[00:10:49] Apple did the same thing.

[00:10:50] They hired a whole team of linguists

[00:10:51] and started listening, started coding

[00:10:53] and then you could opt out of being listened to

[00:10:55] if you look far enough.

[00:10:57] But that's how you had to do it.

[00:10:59] Self-learn.

[00:11:00] At that time we couldn't self-learn.

[00:11:02] Now we can have these models self-learn

[00:11:03] and get better by themselves

[00:11:05] which is really, really cool.

[00:11:06] And that's what deep learning did for us.

[00:11:08] That's what these deep neural nets did for us.

[00:11:10] Yeah, and I just want to specify

[00:11:12] what that means with self-learn

[00:11:13] that it doesn't recognize that

[00:11:16] get me my mail means get me my mail.

[00:11:20] It just knows that the phrase get me my mail

[00:11:23] or I want my mail or where's my mail

[00:11:25] these all belong to the same category.

[00:11:28] It separates that out from get me McDonald's.

[00:11:31] It says to itself, oh,

[00:11:33] there are all these statements

[00:11:35] are still in the get me category

[00:11:37] but in one case there's a,

[00:11:39] they're in two different contexts.

[00:11:40] It doesn't know what mail is.

[00:11:41] It doesn't know what McDonald's is.

[00:11:43] It's the supervised learning later

[00:11:45] that kind of teaches what actions

[00:11:46] you should take based on these different categories.

[00:11:49] Right, that's right.

[00:11:51] That's right.

[00:11:52] So look, getting back to GBT,

[00:11:55] look, GBT is overhyped in a way

[00:11:58] like all new technologies are only

[00:12:00] because it's the first time the public

[00:12:02] got to play with AI.

[00:12:03] Now you and I being intact,

[00:12:05] we've had our various brushes

[00:12:07] with all kinds of AI over decades

[00:12:09] and it makes some progress

[00:12:11] and you use it and it just becomes part of your work.

[00:12:13] Well, all of a sudden the public

[00:12:15] got to type something in

[00:12:16] and it talked back to you

[00:12:17] and you go, oh my goodness.

[00:12:19] Now if you remember in the 60s,

[00:12:21] you know, we had a system from MIT.

[00:12:24] It was Eliza and Eliza, you typed to it

[00:12:27] and it actually come back

[00:12:29] and there's a huge rules engine

[00:12:30] but it was really pretty darn good.

[00:12:32] It was the first instantiation of

[00:12:34] sort of a chatbot that you could talk to.

[00:12:36] And I just want to say

[00:12:37] that the professor that came up with that,

[00:12:38] so Eliza was like an AI therapist

[00:12:41] but it was very rules-based.

[00:12:42] Like if the word contains the word mother

[00:12:47] right back, tell me more about your mother

[00:12:49] and I think it's Joseph Weissenbaum

[00:12:52] professor if I remember correctly.

[00:12:54] And when he was out of the office,

[00:12:57] his secretary was playing with Eliza.

[00:12:58] He walks back in and the secretary says,

[00:13:00] can you stay out for a few more minutes

[00:13:03] as it's getting private?

[00:13:04] No, really, really fascinating

[00:13:09] and so again most of the public

[00:13:12] doesn't realize this kind of chatting

[00:13:14] with an artificial human

[00:13:16] started all the way back in the 60s

[00:13:18] as an experiment

[00:13:19] and we've been getting a little better

[00:13:20] every year and a little better every year

[00:13:22] and then we had products from Amazon,

[00:13:25] we have products from Google,

[00:13:26] we have products from Microsoft

[00:13:27] and they've all been getting better

[00:13:28] and now this has gotten to the point

[00:13:31] where it is almost indistinguishable

[00:13:34] from a human almost.

[00:13:36] You might be able to tell

[00:13:37] but only because you can goat the system

[00:13:39] and it'll kind of give it away

[00:13:40] but generally speaking

[00:13:42] if you didn't know

[00:13:43] you'd say this is very human-like

[00:13:45] and it's so human-like

[00:13:46] because instead of learning

[00:13:48] and you know this

[00:13:49] but instead of learning words

[00:13:51] like we used to do before Transformer

[00:13:53] it learns phrases

[00:13:54] and because it learns phrases

[00:13:56] it just sounds a lot more human

[00:13:57] and it can put together phrases

[00:13:59] that mean something rather than

[00:14:00] put together a bunch of words

[00:14:01] that's just a bunch of junk, right?

[00:14:03] And by the way, that

[00:14:04] hats off to Google in 2017

[00:14:06] who figured out

[00:14:07] if I want to do translations

[00:14:08] to French from English

[00:14:10] I can't just keep translating word by word

[00:14:12] because in France

[00:14:13] that didn't mean anything

[00:14:15] I mean you could figure it out

[00:14:16] but the sentence structure was all wrong

[00:14:18] then they said why don't we just

[00:14:19] learn a whole sentence

[00:14:20] and then you learn a whole sentence

[00:14:21] and then you could formulate sentences

[00:14:23] out of what you said in English

[00:14:25] rather than formulate word by word by word

[00:14:27] and that was the beginning

[00:14:29] of the entire Transformer Revolution

[00:14:31] which again took us another step better

[00:14:33] and now we've got chat GPT

[00:14:35] is everybody over-hyping it?

[00:14:36] Yes

[00:14:37] but will it impact most everything we do?

[00:14:40] Sure

[00:14:41] the way Excel does

[00:14:43] I mean we don't think

[00:14:44] of Excel

[00:14:45] as anything more than an interesting tool today

[00:14:48] but it changed everything we ever did in finance

[00:14:51] period all of us

[00:14:52] Right

[00:14:53] and it didn't replace the entire accounting industry

[00:14:55] in fact there's more accounts than ever

[00:14:57] it didn't replace anything

[00:14:59] in fact because it made us so much more efficient

[00:15:02] as a business

[00:15:03] we were able to build bigger businesses faster

[00:15:06] hire more people

[00:15:07] and the economy developed

[00:15:09] not all because of Excel

[00:15:10] but you know Excel had its effect

[00:15:11] just like many other productivity

[00:15:13] enhancing products

[00:15:14] This is the point

[00:15:15] That's the point

[00:15:16] and the point here is

[00:15:17] people go, oh my job

[00:15:19] I've never seen anything do this

[00:15:20] stop

[00:15:21] just stop

[00:15:22] since the invention of the wheel

[00:15:24] all of these inventions

[00:15:25] improve human productivity

[00:15:26] and when you improve human productivity

[00:15:28] you get more dollars of output per hour

[00:15:31] which the US has been the best at

[00:15:33] probably in the world

[00:15:34] more dollars of output per hour per person

[00:15:36] when you do that

[00:15:37] the companies get bigger

[00:15:38] and I know someone's going to say

[00:15:39] oh only the people at the top make the money etc

[00:15:41] but just trust me

[00:15:42] the companies get bigger

[00:15:43] the GDP gets bigger

[00:15:44] you end up having more money to spend

[00:15:46] on vacations or products

[00:15:48] or consumer goods or whatever

[00:15:49] and around the world goes

[00:15:51] in a very positive way

[00:15:52] the only way today

[00:15:54] you're going to double the size of your company

[00:15:56] because you can't hire a double the number of people

[00:15:58] there are no more people to hire

[00:15:59] we're 10 million behind right

[00:16:01] only way you're going to do it

[00:16:02] is get twice as productive

[00:16:03] as you were three years ago

[00:16:05] and AI is going to help you be twice as productive

[00:16:08] in marketing I can now write a blog post

[00:16:10] in ten now

[00:16:12] I'm going to probably spend

[00:16:13] a half an hour editing it

[00:16:14] but it used to take me hours to write it

[00:16:16] and now I can probably write it in one minute

[00:16:18] and spend a half an hour editing

[00:16:19] and I'm done

[00:16:20] so I may be 80% more efficient

[00:16:22] in writing a blog post

[00:16:23] than I've ever been in my life

[00:16:27] you know I was talking to Matt Barry

[00:16:41] who's the CEO of Freelancer.com

[00:16:43] they have 60 million freelancers on their platform

[00:16:45] he said A the number of freelancers is shooting up

[00:16:47] which is why kind of the unemployment data

[00:16:49] is so skewed

[00:16:50] but it's great

[00:16:52] like let's say you're a logo designer

[00:16:53] now instead of doing one logo a week

[00:16:55] you could do ten logos a day

[00:16:57] because of midrity

[00:16:59] but the companies still need

[00:17:01] someone with a design aesthetic

[00:17:03] they still need a human to kind of manage the product

[00:17:06] you're not going to just be the CEO of company

[00:17:09] and you know design your logo

[00:17:12] because you still don't know if it's good or bad

[00:17:14] like you still need to hire the logo designer

[00:17:16] I think you're right on

[00:17:18] is that what we're going to find

[00:17:20] is that we are going to

[00:17:22] we're going to make a few logos

[00:17:24] we're going to make a few graphics ourselves

[00:17:26] and then we're going to go

[00:17:28] I'm not actually a graphic person

[00:17:30] now we did democratize

[00:17:32] access to tools that allow you

[00:17:34] to generate graphics that you never generated before

[00:17:36] or to generate music

[00:17:38] or to generate video

[00:17:40] all these things this is amazing

[00:17:42] but in the end you're going to hand it to a professional

[00:17:44] but instead of them taking three weeks

[00:17:46] they're going to take an hour

[00:17:48] and they're going to do it at a quarter of the cost

[00:17:50] and they're going to give you ten things to look at

[00:17:52] and you're going to be out of that loop in an hour

[00:17:54] or a day instead of what used to take weeks

[00:17:56] and that's what I'm finding

[00:17:58] I'm designing the next PowerPoint or Keynote or whatever

[00:18:00] you know I am getting

[00:18:02] all this material that I

[00:18:04] didn't have that used to take weeks to generate

[00:18:06] that can be generated in minutes or hours

[00:18:08] what I tell people is exactly this

[00:18:10] that don't think of AI

[00:18:12] as the terminator

[00:18:14] think of AI as your new assistant

[00:18:16] and whatever you want to do

[00:18:18] whether you're researching for a podcast

[00:18:20] like this one or whether you're

[00:18:22] outlining a book

[00:18:24] thinking of business ideas

[00:18:26] AI can assist you with the right prompts

[00:18:28] with the right questions

[00:18:30] I don't think you're going to write the most successful novel in the world

[00:18:32] just using AI because that needs human experience

[00:18:34] and going beyond

[00:18:36] the human frontier to write something no one's ever read before

[00:18:38] same thing with music

[00:18:40] same thing with a lot of kinds of art

[00:18:42] or not we'll see

[00:18:44] but I also don't think

[00:18:46] people think oh well this is AI now

[00:18:48] I don't think we're going to see

[00:18:50] exponential improvements

[00:18:52] they've already crunched the trillion

[00:18:54] tokens and everything written in all of history

[00:18:56] and this is what they have

[00:18:58] and it still requires several years of human

[00:19:00] supervised learning so it's not going to be like

[00:19:02] oh it can write

[00:19:04] a bad novel now so next year it's going to write

[00:19:06] the most amazing novel in the world

[00:19:08] it's just going to have incremental improvements

[00:19:10] the technology is out of the box now

[00:19:12] now is where we have this computer vision

[00:19:14] that this has been a solved problem for at least 20 years

[00:19:16] and it's not gotten

[00:19:18] that much better

[00:19:20] we're still doing the captures

[00:19:22] where we identify these are bicycles

[00:19:24] and these are street signs and then you have to supervise it

[00:19:26] it's supervised learning

[00:19:28] now our rec rates

[00:19:30] on images has now approached

[00:19:32] 98-99% it's actually

[00:19:34] better than humans

[00:19:36] like a wolf versus or a fox

[00:19:38] versus a dog for example

[00:19:40] a coyote versus a dog

[00:19:42] the AI is better than we are necessarily

[00:19:44] but what's interesting is past about

[00:19:46] 2015 it stopped getting better

[00:19:48] we got to 98-99%

[00:19:50] in image rec and that turns out

[00:19:52] to be better than humans are and it just really

[00:19:54] didn't get any better than that because already

[00:19:56] we have unlimited depth of neural nets

[00:19:58] unlimited training and eventually

[00:20:00] just run out of stuff to train right there's

[00:20:02] I have no more data I have no more

[00:20:04] supervised information to train you on

[00:20:06] that doesn't start to corrupt

[00:20:08] the data and this is another thing

[00:20:10] I would say

[00:20:12] two things one AI makes these jumps

[00:20:14] and you're absolutely right on

[00:20:16] GPT. GPT made a jump in large language models

[00:20:18] and now there will be a bunch of large language models

[00:20:20] that work faster or better

[00:20:22] work better or whatever

[00:20:24] and that jump has been made and now there will be

[00:20:26] tiny little incremental improvements but if you go from

[00:20:28] a trillion tokens to ten trillion tokens

[00:20:30] no it's not going to get that much

[00:20:32] better you've asymptoted

[00:20:34] all you can get out of that database

[00:20:36] I'm going to use that as an insult

[00:20:38] you asymptoted your way out of that situation

[00:20:40] you jerk like

[00:20:42] well I mean that's all there is right

[00:20:44] they kind of did this and leveled off

[00:20:46] nothing wrong with that the other thing I'm going to give you

[00:20:48] a term I don't know if you like it or not

[00:20:50] and I kind of stole it but I

[00:20:52] so I used to say augmented intelligence what's AI

[00:20:54] what's augmented intelligence you got this

[00:20:56] assistant actually

[00:20:58] it's amplified intelligence because

[00:21:00] I can take my one brain power

[00:21:02] now and essentially

[00:21:04] be the equivalent of 10 or 100 brain

[00:21:06] powers right in almost any

[00:21:08] field whether it's medical

[00:21:10] whether it's writing a novel whether it's writing

[00:21:12] marketing if they whether it's coding all of a

[00:21:14] sudden I can code 10 times faster than I was

[00:21:16] able to code now I still got

[00:21:18] to debug it and it's still got problems and so

[00:21:20] I still needs me but I'm the equivalent

[00:21:22] of many many many brain

[00:21:24] power now instead of one brain power

[00:21:26] this is incredible right

[00:21:28] and everybody's going to have more than one

[00:21:30] brain power now that has an interesting

[00:21:32] outcome it says

[00:21:34] that IQ overall

[00:21:36] will get somewhat depreciated because

[00:21:38] everyone can have

[00:21:40] 3 5 10 20 100

[00:21:42] brain power and

[00:21:44] eq in collaboration will become

[00:21:46] more important and because

[00:21:48] if you've got all this brain

[00:21:50] power you're going off in a direction but you didn't collaborate with anyone else

[00:21:52] you could be down

[00:21:54] in the wrong direction right so

[00:21:56] collaboration between team members

[00:21:58] is going to be really really

[00:22:00] appreciated and IQ

[00:22:02] anyone can use these tools

[00:22:04] right once you learn to use the tools you got the IQ of the tool which is

[00:22:08] really high so I wonder

[00:22:10] that's an interesting question I wonder if

[00:22:12] per industry it'll be

[00:22:14] possible to measure the difference

[00:22:16] between someone using AI as an

[00:22:18] assistant versus someone not using AI as an assistant

[00:22:20] so here's an example like let's say

[00:22:22] I'm going to take a random one like

[00:22:24] tennis so let's say

[00:22:26] an AI assistant can watch

[00:22:28] you play tennis see the arc

[00:22:30] of your serve you know when you swing

[00:22:32] and study all your

[00:22:34] health and your workout regime

[00:22:36] and your training and your opponent

[00:22:38] and your games and it gives you suggestions on

[00:22:40] how to train, how to swing, what equipment

[00:22:42] to use, what to eat, how to play

[00:22:44] against an opponent and so on I wonder

[00:22:46] how much better the AI helped

[00:22:48] person will be than the non-AI

[00:22:50] help person and if that could be done per

[00:22:52] industry like writing, music

[00:22:54] chess, business and so on

[00:22:56] well that's a great question

[00:22:58] so I think it depends on what you've got the AI

[00:23:00] doing so in this case we're asking the AI

[00:23:02] to be my tennis trainer as opposed

[00:23:04] to a professional trainer right and then

[00:23:06] you get to who was the professional trainer and how much

[00:23:08] experience do they do and how good are they interacting

[00:23:10] with me and all of those other things but

[00:23:12] augment it with a human trainer as well

[00:23:14] oh it's got to be better I mean

[00:23:16] as long as it can be multimodal that is

[00:23:18] as long as I can give feedback and it says

[00:23:20] oh I saw your stroke and it's this

[00:23:22] it's a little bit like can

[00:23:24] I

[00:23:26] evaluate ECGs better with

[00:23:28] AI than I can move cardiologists

[00:23:30] well of course I can that was solved five years

[00:23:32] ago like we know we can because

[00:23:34] the cardiologist will miss things that are ever

[00:23:36] so nuanced that AI will see

[00:23:38] and this is true with X-ray readings right

[00:23:40] I mean you know you can the AI

[00:23:42] will see things that the human eye just

[00:23:44] won't pick up yeah in fact they passed

[00:23:46] a law then that a human has to

[00:23:48] be the one to tell you the news

[00:23:50] yes human has to be one to tell you the news

[00:23:52] even though you're using these systems that know better

[00:23:54] so then the human evaluates that human doctor

[00:23:56] says okay it's your radiologist as well

[00:23:58] okay well I'll agree with this or I disagree

[00:24:00] with it or whatever the AI is right

[00:24:02] now the problem with AI is

[00:24:04] a bunch of hidden layers right so when you've got

[00:24:06] these neural nets you got a bunch of hidden layers

[00:24:08] and people get angry well that's not explainable

[00:24:10] here's my explanation

[00:24:12] of why it's not explainable like when I

[00:24:14] look at you I can say oh

[00:24:16] I know you curly hair glasses

[00:24:18] probably has the headphones on as a microphone

[00:24:20] I get it right like

[00:24:22] I know that right I know how to recognize you

[00:24:24] but AI will recognize you

[00:24:26] in a thousand different ways and there isn't

[00:24:28] an English word

[00:24:30] for each of those methods all those hidden level

[00:24:32] layers they don't mean to hide it it's just

[00:24:34] there's no way to explain

[00:24:36] what that was I saw a little thing here

[00:24:38] there's no explain so

[00:24:40] so that's the that that's sort of the

[00:24:42] interesting thing is that

[00:24:44] you want AI to tell you exactly

[00:24:46] what it found and it says I can't

[00:24:48] but after looking at a million x-rays

[00:24:50] I know this person has cancer I have

[00:24:52] no way to tell you why I know that or

[00:24:54] even maybe where I see it

[00:24:56] it's just this is a pattern I've recognized

[00:24:58] before and the AI

[00:25:00] is unlikely to be wrong in those cases

[00:25:02] if it was well trained right now

[00:25:04] let me play the devil's advocate a little bit

[00:25:06] because I've spoken to a lot of people

[00:25:08] and a lot of smart people and

[00:25:10] I've almost been disturbed by how

[00:25:12] negative they are and just as an example

[00:25:14] I was in New York City the other day

[00:25:16] and I had dinner with a bunch of

[00:25:18] hardcore scientists like these were

[00:25:20] professors, well known physicists

[00:25:22] well known chemists, biologists

[00:25:24] a Pulitzer Prize winning science

[00:25:26] writer and other science writers

[00:25:28] and they were universally

[00:25:30] pessimistic about almost every

[00:25:32] technology but particularly AI

[00:25:34] like oh you know AI is going to

[00:25:36] basically turn the guns on the humans

[00:25:38] and decide to take over

[00:25:40] the world and all this other stuff

[00:25:42] hey AI is going to make viruses

[00:25:44] that kill everybody like

[00:25:46] okay I get it some people

[00:25:48] tend to be more negative than

[00:25:50] positive but

[00:25:52] what are the negative aspects of it

[00:25:54] like okay let's take another one I had a conversation with

[00:25:56] Vernon Reed who was the

[00:25:58] head of the band Living Color

[00:26:00] had this on cult of personality in the 80s

[00:26:02] one of my favorite songs and

[00:26:04] he was like what are you going to do when

[00:26:06] an AI version of your

[00:26:08] father's voice calls you and says

[00:26:10] you know son I'm stuck in the middle of nowhere

[00:26:12] I need you to wire this money right away

[00:26:14] please help me like is there going to be

[00:26:16] more what kind of problems are going to result

[00:26:18] sure sure great great

[00:26:20] great question a great conversation piece

[00:26:22] so first of all everything

[00:26:24] AI is doing today in terms of deep fakes

[00:26:26] from voice to visual has been

[00:26:28] available in Hollywood for 25 years

[00:26:30] right this is not like we know that

[00:26:32] Hollywood has been able to de-age people

[00:26:34] or put people in movies that have been dead

[00:26:36] and all kinds of things and they're

[00:26:38] incredibly convincing and put people

[00:26:40] into you know change their face

[00:26:42] and do all that right in fact create

[00:26:44] a lot of nothing but as Pixar has been

[00:26:46] doing for 25 years and

[00:26:48] and this is a really fascinating

[00:26:50] thing all this technology

[00:26:52] is doing is democratizing access

[00:26:54] to those tools so those tools

[00:26:56] you know used to render

[00:26:58] like one frame a minute and take

[00:27:00] six months to render right they were

[00:27:02] hugely you know painful

[00:27:04] to do but the models for

[00:27:06] them were understood they weren't the same as

[00:27:08] AI models are today they were rules

[00:27:10] based but nevertheless Hollywood's had these

[00:27:12] tools but we knew that Hollywood

[00:27:14] wouldn't use them to call you

[00:27:16] and try to get your money right we knew

[00:27:18] that so we trusted that so

[00:27:20] none of this is new deep fakes are not

[00:27:22] new we could make Obama say anything we wanted

[00:27:24] to 20 years ago however it required

[00:27:26] Hollywood tools and millions of dollars

[00:27:28] so we did it now

[00:27:30] we democratized access

[00:27:32] even to this

[00:27:34] and by democratizing access it means

[00:27:36] we can use it for free or a dollar

[00:27:38] or two and so can bad guys

[00:27:40] right so

[00:27:42] but this has been true with every tool in the

[00:27:44] world that has come forth forever

[00:27:46] right bad guys use them

[00:27:48] bad guys use everything there's bad guys

[00:27:50] who use excel I mean it just

[00:27:52] is what it is right they use banks

[00:27:54] they use the banking system so here's another

[00:27:56] set of things and and it is

[00:27:58] true that already

[00:28:00] people are getting calls from their mom

[00:28:02] or their grandmother and it sounds like exactly like

[00:28:04] them and so please wire the money and

[00:28:06] they think it's them etc etc so

[00:28:08] we're gonna have to be more diligent on what

[00:28:10] we believe because you will

[00:28:12] no longer believe be able to believe

[00:28:14] that

[00:28:16] watching that president in that park

[00:28:18] exchanging money with that person

[00:28:20] actually happened you have no way

[00:28:22] in fact you may have no way to tell whether it

[00:28:24] actually happened or not and for all of history

[00:28:26] we've always believed it happened

[00:28:28] unless Hollywood told us otherwise

[00:28:30] because they've all they've told us otherwise for decades

[00:28:32] then that's not new right

[00:28:34] but now this isn't Hollywood it's just

[00:28:36] Joe Schmo down the street some kid

[00:28:38] who generates it right so

[00:28:40] we're just going to be more thoughtful about what we believe

[00:28:42] and what we don't believe and you will not be able to believe

[00:28:44] every video that you see let me say one other thing

[00:28:46] people have been photoshopping

[00:28:48] you know things together at least

[00:28:50] in still frame I don't know

[00:28:52] for three decades right so

[00:28:54] so fakes deep fakes

[00:28:56] on Photoshop have been around but you needed

[00:28:58] to be a good Photoshop jockey

[00:29:00] to do it well enough to hide the fact

[00:29:02] that you put the face on someone else's body

[00:29:04] and you know and maybe you hid that

[00:29:06] with a necklace and all these tricks right

[00:29:08] so that's been done in Photoshop for 30 years

[00:29:10] but you needed to know Photoshop

[00:29:12] which wasn't that except for $100

[00:29:14] you could learn Photoshop

[00:29:16] now people can do it by just typing

[00:29:18] text put this hat on this body

[00:29:20] and there it is

[00:29:22] you go okay well

[00:29:24] now I put them in a place they weren't

[00:29:26] change the background put them in front of an oil plant

[00:29:28] put them with money in their hand

[00:29:30] you can do that with anything right

[00:29:32] so that's sort of democratize the access to it

[00:29:34] so I think this is the real problem

[00:29:36] like this is the real negative thing if you want to find something

[00:29:38] negative about AI is that

[00:29:40] we're going to be dealt with a lot more

[00:29:42] let's call them you know scam

[00:29:44] fake news phishing attacks

[00:29:46] scams and it's going to be harder and harder

[00:29:48] to verify whether something's

[00:29:50] true so verification

[00:29:52] is the key if someone claiming

[00:29:54] to be your mother calls you and needs money desperately

[00:29:56] you just got to verify somehow that this

[00:29:58] is a true thing

[00:30:00] call your mother back your actual mother

[00:30:02] on her actual phone

[00:30:04] this is not hard it takes an extra minute

[00:30:06] right and that's

[00:30:08] fine that is not a

[00:30:10] bad thing the other thing

[00:30:12] is throughout history

[00:30:14] when technology

[00:30:16] comes forward new technology whatever it is

[00:30:18] you notice and I love scientists

[00:30:20] we need scientists you know it's usually scientists

[00:30:22] oh it's the end of humanity

[00:30:24] oh it's going to be all over

[00:30:26] okay look

[00:30:28] we have a language model that is doing

[00:30:30] for text and language

[00:30:32] what the calculator and excel

[00:30:34] did for math that's all

[00:30:36] and we solved it in math decades ago

[00:30:38] and now we're doing it for language which is a

[00:30:40] much bigger problem do not

[00:30:42] hook the language model to the nuclear

[00:30:44] arsenal okay this is a bad idea

[00:30:46] right and and and by

[00:30:48] the way we've had the ability

[00:30:50] because I was once

[00:30:52] offered a job to work on this with at MIT

[00:30:54] Lincoln labs we've had the ability

[00:30:56] to recognize using AI

[00:30:58] which objects in space were junk

[00:31:00] and which objects in space were nuclear

[00:31:02] missiles heading towards us we've had that

[00:31:04] tech for 30 years so

[00:31:06] never then did we hook up

[00:31:08] the AI to you know make

[00:31:10] the decisions hey

[00:31:12] in fact we signed a treaty with Russia

[00:31:14] back in the 80s to not use

[00:31:16] any form of AI to automatically launch weapons

[00:31:18] to give it to people

[00:31:20] who would then make the decision there's actually

[00:31:22] a treaty against this against using

[00:31:24] AI and hooking it up directly because we knew

[00:31:26] in the 80s even with the AI we had then

[00:31:28] which was very simple that this

[00:31:30] was a bad thing right this

[00:31:32] right like so so it's now like

[00:31:34] suddenly we're going to change our mind to say

[00:31:36] you know what I want to kick back and have

[00:31:38] doughnuts I'm gonna let AI make these decisions

[00:31:40] for me it's not like magically we're going to

[00:31:42] give AI the power to do something they

[00:31:44] could have done 30 years ago

[00:31:46] right that's right no nobody's going to hook

[00:31:48] it to the nuclear arsenal nobody's going to hook it to crazy things

[00:31:50] everybody relax it is

[00:31:52] not sentient it is not dangerous

[00:31:54] it's just a large language model

[00:31:56] just like we have large math models and

[00:31:58] by the way a large language model

[00:32:00] is actually a large math model

[00:32:02] it's only math yeah I mean

[00:32:04] in the end all we're doing is we're signing

[00:32:06] probabilities to neurons

[00:32:08] right to neural nets to say well the probability

[00:32:10] to put this word after that one

[00:32:12] based on what they you know ask me

[00:32:14] is here it's easy to

[00:32:16] well use

[00:32:18] yeah and that's all it's doing is

[00:32:20] forming sentences everybody relax right the scientists

[00:32:22] god bless or wrong

[00:32:24] now you got a lot out there going oh

[00:32:26] many eyes going to kill everyone then he starts his own

[00:32:28] company so I think

[00:32:30] that's a little disingenuous

[00:32:32] it was all positioning

[00:32:34] no I agree whatever like the one of these guys

[00:32:36] says something even even like

[00:32:38] Warren Buffett for his entire career

[00:32:40] if he says you know there might be a chance

[00:32:42] terrorists could attack of

[00:32:44] the Superbowl with nuclear bombs

[00:32:46] what does he do next he's an insurance company

[00:32:48] and he tells insurance to every Superbowl

[00:32:50] yes yes he does

[00:32:52] so you always have to look at what the agenda

[00:32:54] of these guys are and

[00:32:56] godspeed are the most innovative people

[00:32:58] around in some cases but

[00:33:00] Elon Musk certainly created a lot of innovation

[00:33:02] but why do you think

[00:33:04] that Google engineer thought the AI

[00:33:06] he was working on was sentient

[00:33:08] presumably he's tech savvy he knows everything

[00:33:10] any other tech person does

[00:33:12] why would he think it was sentient

[00:33:14] I could only guess

[00:33:16] that Elon's going to hang me for I can only guess

[00:33:18] that it was kind of his subconsciously

[00:33:20] his time to get in the limelight

[00:33:22] and today you're not in the limelight if you're

[00:33:24] going listen it's just a freaking

[00:33:26] math model everyone calm down

[00:33:28] it's oh it's going to kill us put you on

[00:33:30] every podcast right so

[00:33:32] I'm here clearly stating

[00:33:34] because I understand the underlying math and so do

[00:33:36] you it's freaking math everyone calm

[00:33:38] down don't hook it to the nuclear arsenal

[00:33:40] we're fine but

[00:33:42] if you're him and you're one of

[00:33:44] lots and lots of important people at

[00:33:46] the mind in Google AI and everything else

[00:33:48] you know you're not getting any press by

[00:33:50] saying this is really cool in fact

[00:33:52] you're not getting anything so it's better for you to

[00:33:54] leave and say I have interacted

[00:33:56] with this thing and I think it's actually sentient

[00:33:58] okay well now you get all the press

[00:34:00] in the world and now you're really really important

[00:34:02] right but it's just BS

[00:34:04] it's just math if you

[00:34:06] understand the math it can't

[00:34:08] be sentient it does not know

[00:34:10] what it's saying it learned

[00:34:12] from trillions of phrases

[00:34:14] and it put them back together for you in

[00:34:16] some random way but you know thoughtful

[00:34:18] way from novels

[00:34:20] and from truth and from fiction and from

[00:34:22] everything else with a huge rules engine around

[00:34:24] it again that took a year and a half to put together

[00:34:26] everyone calm down right

[00:34:31] it's sort of like a simple

[00:34:44] way to think about it is the Google search box

[00:34:46] if you type in the letters s a n

[00:34:48] and you live in San Francisco

[00:34:50] Google is smart enough to know that Francisco

[00:34:52] is probably the next word you're going to type

[00:34:54] it's just using the same statistics

[00:34:56] but like you said it went from words to phrases

[00:34:58] you have a much more sophisticated

[00:35:00] and larger model

[00:35:02] now what about cybersecurity

[00:35:04] like what's you know obviously

[00:35:06] AI is not going to be used by the bad guys

[00:35:08] to figure out where there

[00:35:10] might be holes in operating systems

[00:35:12] and so on that they can exploit

[00:35:14] so here's the interesting thing since

[00:35:16] the advent of ransomware

[00:35:18] everything in cybersecurity has changed

[00:35:20] so it used to be bad guys got

[00:35:22] into your network and messed it up but

[00:35:24] because they hated you or because they could

[00:35:26] or because then ransomware came around

[00:35:28] and it was like I can charge a hundred

[00:35:30] thousand dollars to not

[00:35:32] reveal their data on the web or just

[00:35:34] not steal whatever to give them back

[00:35:36] their data give max whatever

[00:35:38] okay then we started encrypting our data

[00:35:40] and that meant that people getting

[00:35:42] into our network maybe they could do some

[00:35:44] network damage but they couldn't do anything

[00:35:46] with the data because all the databases have become encrypted

[00:35:48] except

[00:35:50] our applications all of them know how

[00:35:52] to decrypt the data so now you want access

[00:35:54] to the applications because then that's

[00:35:56] how you get the data so then they go

[00:35:58] well I'd have to come in the front door

[00:36:00] well it turns out that's freaking easy because

[00:36:02] passwords and IDs are

[00:36:04] dumb people use their kids names

[00:36:06] and one two three four

[00:36:08] and use the word password is the most common password

[00:36:10] right so so

[00:36:12] what happened is there's an entire network

[00:36:14] now of

[00:36:16] of identity access brokers

[00:36:18] and what they do is if you're

[00:36:20] a ransomware person you go to

[00:36:22] an identity access broker and say

[00:36:24] I want access to

[00:36:26] someone at Bank of America I'm making up some

[00:36:28] some teller so someone who has access to accounts

[00:36:30] or whatever you want to do right that comes

[00:36:32] in through the front door and

[00:36:34] they go and get you that access

[00:36:36] including MFA

[00:36:38] so all MFA all traditional MFA

[00:36:40] has been hacked what's MFA

[00:36:42] oh multi-factor authentication so the codes

[00:36:44] that come on your phone yeah ridiculously

[00:36:46] easy to hack they're easy and I can

[00:36:48] I can show you the social hacks right now

[00:36:50] I won't tell everyone how we do it technically

[00:36:52] but technically is harder

[00:36:54] but actually pretty easy like in minutes

[00:36:56] because I have your cell phone I can

[00:36:58] certainly get all your codes this afternoon

[00:37:00] that wouldn't be hard okay

[00:37:02] but the easiest way to get MFA

[00:37:04] is actually simply a social hack

[00:37:06] and social engineering is

[00:37:08] very easy so let's say

[00:37:10] you're at a company

[00:37:12] but you're at a big company not a small company

[00:37:14] okay and the big company you're one of

[00:37:16] 50,000 employees

[00:37:18] IT calls you

[00:37:20] says hi this is IT this is Jim from IT

[00:37:22] or Bill from IT

[00:37:24] James we've you know we

[00:37:26] we think we've got to break into your account

[00:37:28] what I'm going to do is I'm going to

[00:37:30] initiate a login and password

[00:37:32] that's going to send you a code

[00:37:34] I don't have access to that code

[00:37:36] so I need you to share your code once by the way

[00:37:38] just to validate it your employee number is this

[00:37:40] your social security number is this you go oh okay

[00:37:42] you're on the phone with them

[00:37:44] right you get your code you say it's

[00:37:46] one two three four whatever it is right

[00:37:48] they then put that in they now

[00:37:50] have access to your account they say you know what

[00:37:52] we're in we're in together looks

[00:37:54] like everything's fine thank you very much goodbye

[00:37:56] you hang up

[00:37:58] they now have access to that data

[00:38:00] they're sucking out data take some 20

[00:38:02] minutes they've got everything they want

[00:38:04] you handed them the

[00:38:06] multi-factor authentication from your phone or from RSA

[00:38:08] card or whatever it was

[00:38:10] and now they're going to hold the company ransom

[00:38:12] I know in one case 22 million dollars recently

[00:38:14] so oh my god did they pay

[00:38:16] did that company pay oh yeah oh yes

[00:38:18] what was the data

[00:38:20] roughly you don't have to say exactly it had to do

[00:38:22] with running machines

[00:38:24] and what they did is they

[00:38:26] essentially went in and

[00:38:28] changed all the data

[00:38:30] and they didn't have a backup for several weeks and they wouldn't be able to

[00:38:32] operate the machines it basically shut them down

[00:38:34] and so it was cheaper to pay the 22

[00:38:36] million and bring the place back up

[00:38:38] I have example after example

[00:38:40] by the way everybody pays

[00:38:42] and almost no one knows they don't publicize it

[00:38:44] right they try to just bury it

[00:38:46] this is true with universities it's true with hospitals

[00:38:48] we've you know

[00:38:50] every hospital pays they cannot let that

[00:38:52] patient information out you know that's it

[00:38:54] so then you say I got to close

[00:38:56] the front door so one of my companies

[00:38:58] that you're not an investor but

[00:39:00] it's a great company it's called token ring

[00:39:02] token ring is literally

[00:39:04] I don't have one on but it looks like this it's a ring

[00:39:06] that you put on your finger and when you put it

[00:39:08] on it's tied to your fingerprints so you've registered

[00:39:10] your applications with this ring

[00:39:12] okay that ring now

[00:39:14] only works on your finger no one else's finger

[00:39:16] they could steal it you can lose it

[00:39:18] it doesn't matter right and your bio

[00:39:20] information is stored inside the ring in a secure

[00:39:22] element so if you try to get in there it destroys

[00:39:24] the information you know in the movies

[00:39:26] they're going to just cut off your finger and use it

[00:39:28] well yes in the movies

[00:39:30] but it turns out we're not trying to

[00:39:32] solve for people cutting off your arm that's a different

[00:39:34] problem right that's a that's a different

[00:39:36] level of crime as you know most of this ransomware

[00:39:38] happens for people far far away

[00:39:40] in Russia and China

[00:39:42] and in North Korea so this is fascinating

[00:39:44] because now

[00:39:46] your access is tied to that ring

[00:39:48] and it goes behind

[00:39:50] your application server to something called a

[00:39:52] phyto 2 server to validate that you are

[00:39:54] you 128 bit encryption

[00:39:56] happens in one second so we can

[00:39:58] we can't decode that today and that

[00:40:00] speed and it's different for every application

[00:40:02] that validation so it

[00:40:04] it makes your access

[00:40:06] 100% foolproof and

[00:40:08] no one can call you and say give me

[00:40:10] your number you go I don't have a number

[00:40:12] what are you talking about have a ring there's

[00:40:14] there's no information on the ring there's no serial

[00:40:16] number I can't give you anything and they

[00:40:18] hang up right they go well you're not

[00:40:20] helpful I can't do anything right

[00:40:22] so so the solution is to close

[00:40:24] the front door the solution is with

[00:40:26] I think wearable biometrics

[00:40:28] and anyway so

[00:40:30] that's one of the other companies I have because

[00:40:32] I think cyber security has to be solved

[00:40:34] and in this case AI can't beat it either

[00:40:36] there's nothing AI can do

[00:40:38] because they can fish you

[00:40:40] all you want but they can't get your ring

[00:40:42] doesn't matter they can't get

[00:40:44] the ring and get access you know

[00:40:46] phishing is all about hey this is easy

[00:40:48] I'm gonna steal your password and your login ID

[00:40:50] but if they can't steal your password

[00:40:52] if they can steal even your password and login ID

[00:40:54] but they don't have the ring they still can't get it

[00:40:56] will quantum decryption

[00:40:58] eventually be there for decrypting

[00:41:00] are we decades away from that

[00:41:02] so as you know quantum computers are coming

[00:41:04] along and we're looking at things that are getting

[00:41:06] into the hundreds and then thousands of qubits

[00:41:08] now so this is really fascinating however

[00:41:10] we're talking about two or three

[00:41:12] quantum computers in the entire world

[00:41:14] that you know live at google and

[00:41:16] IBM right so like there's no one has

[00:41:18] access to these things right now so I

[00:41:20] think we're at least a decade away from

[00:41:22] bad guys having maybe two decades

[00:41:24] away from bad guys having access

[00:41:26] to a billion dollar quantum computer

[00:41:28] right this this access is highly

[00:41:30] limited and it's for researchers

[00:41:32] right now and these things are only doing

[00:41:34] sort of one thing a quantum computer

[00:41:36] could certainly

[00:41:38] decrypt a hundred and twenty eight

[00:41:40] bit encryption if you could do a man in the

[00:41:42] middle attack it could decrypt it

[00:41:44] in a second or less however

[00:41:46] you've already completed your

[00:41:48] transaction and it's gone and the next

[00:41:50] time that you know those encryption keys will be

[00:41:52] different so it still

[00:41:54] may not solve it because that rotates

[00:41:56] around it's not as simple as

[00:41:58] I've decrypted it and now I've got it

[00:42:00] right by the time you've decrypted it

[00:42:02] that the you know they're already logged on

[00:42:04] and they're doing their thing and there's nothing you can do right

[00:42:06] it's it's a little bit late so I think

[00:42:08] it'll still be hard I think it'll be two decades

[00:42:10] before bad people have quantum computers

[00:42:12] and by then we may do you know

[00:42:14] 512 bit 1024 bit

[00:42:16] encryption yeah there are

[00:42:18] algorithms off also for

[00:42:20] when there is a quantum

[00:42:22] encryption or decryption

[00:42:24] but I just don't know how far advanced those

[00:42:26] algorithms are and I don't know how fast

[00:42:28] we can employ them the

[00:42:30] the problem is the speed of the processors right so

[00:42:32] think about the processor

[00:42:34] I have inside a ring that I have to do

[00:42:36] decryption on so I could do 128 bit

[00:42:38] encrypt decrypt but

[00:42:40] if I had to do some quantum encrypt

[00:42:42] decrypt that's just a massive amount

[00:42:44] of processing power that I might not

[00:42:46] have in a bio ring right

[00:42:48] which has a tiny little processor in it

[00:42:50] so that's the problem but

[00:42:52] you know that's a problem 20 years from now

[00:42:54] again no one has access to these things

[00:42:56] but researchers at IBM and Google so

[00:42:58] you can't log on and say I'm using a quantum

[00:43:00] computer as a bad guy in Russia

[00:43:02] you don't have one and Russia

[00:43:04] is not going to develop one right they

[00:43:06] yeah they lost whatever you know

[00:43:08] brainpower they had so and I guess

[00:43:10] the other thing that's very interesting in the AI

[00:43:12] space is how many AI companies

[00:43:14] have started overnight like

[00:43:16] there's like hundreds of thousands of AI

[00:43:18] companies have started in the past few months

[00:43:20] and my own god

[00:43:22] feeling is when I look at these companies

[00:43:24] is that there's nothing special like it's so

[00:43:26] easy now to just

[00:43:28] hook up to the API of chat GPT

[00:43:30] and make

[00:43:32] a pseudo company that anyone else could

[00:43:34] also create overnight so

[00:43:36] on the one hand the price of making an

[00:43:38] AI company has gone to zero

[00:43:40] but on the other hand

[00:43:42] none of these things are worth anything because they're not doing anything

[00:43:44] of value that's right where's

[00:43:46] the mode right do you remember when mobile

[00:43:48] once the iPhone came out and then there had to

[00:43:50] be a hundred thousand mobile first companies

[00:43:52] I'm doing mobile I'm doing and that were

[00:43:54] I don't know easily

[00:43:56] a million app companies that were writing apps for

[00:43:58] iPhone right are there

[00:44:00] like 20 that ever made money some

[00:44:02] number like that it's amazing right so very very

[00:44:04] small number of apps that ever made any money

[00:44:06] yet all these companies got VC funding

[00:44:08] tens of thousands of companies got VC

[00:44:10] funding to do some app for the iPhone

[00:44:12] of which nobody made any money in fact

[00:44:14] all the downloads are

[00:44:16] Facebook, Twitter you know you can just go

[00:44:18] down that you know what they are right there's ten downloads

[00:44:20] that everybody has and then usage on

[00:44:22] their phone goes goes pretty low

[00:44:24] so most of those app companies

[00:44:26] didn't get anywhere and that's

[00:44:28] going to be true with AI right there's the rush to

[00:44:30] everybody's going to do AI but but if

[00:44:32] you can write an AI interface

[00:44:34] in a weekend your mode is

[00:44:36] too small now there's other companies

[00:44:38] you know app answers an example it's

[00:44:40] taking years of AI effort

[00:44:42] to get to where we are yes we're

[00:44:44] also leveraging large language models

[00:44:46] in unique ways but even that

[00:44:48] is another year of

[00:44:50] effort to make that work across

[00:44:52] you know the plethora of mobile

[00:44:54] and web applications it's not

[00:44:56] something you do in a weekend and hook up in a weekend

[00:44:58] and go I'm in the test business now

[00:45:00] there's a few startups that have said oh that's what

[00:45:02] we're doing good luck when do you have

[00:45:04] to test every application with every library

[00:45:06] with every different set of code

[00:45:08] that someone has created turns out it's

[00:45:10] very hard it's really really hard

[00:45:12] and you've got to use visual information

[00:45:14] it's got to be multi-modal and then

[00:45:16] people give up and I've seen this happen

[00:45:18] people start and a month later they're gone

[00:45:20] and yeah you know two months later and they go

[00:45:22] this is really freaking hard I thought it'd be easy

[00:45:24] so so I think the easy ones

[00:45:26] you're right and they probably

[00:45:28] shouldn't get funded and there's a rush to fund them

[00:45:30] and the hard things will be hard

[00:45:32] and some of those are going to succeed

[00:45:34] yeah and again I think just to

[00:45:36] wrap this up just as a mindset

[00:45:38] view a for every

[00:45:40] possible problem where you get scared

[00:45:42] oh AI is going to replace me

[00:45:44] this is going to happen this bad thing

[00:45:46] think of AI as an assistant instead

[00:45:48] like AI will be a very good assistant

[00:45:50] and will make your life better

[00:45:52] and it's not that hard to figure out and start using

[00:45:54] you do it today you could just go to

[00:45:56] you know whatever let's just say you go to

[00:45:58] openai.com chat.openai.com

[00:46:00] sign up for free and start using

[00:46:02] it and you'll see you can use it as

[00:46:04] an assistant for for most of your jobs

[00:46:06] co-pilot that's now been out

[00:46:08] two years from Microsoft by the way

[00:46:10] it's in GitHub right two years we've had

[00:46:12] co-pilot that helps you code

[00:46:14] and it's getting better and better it's based on

[00:46:16] GPT-4 now that is

[00:46:18] making coders latest stats

[00:46:20] 55% more productive

[00:46:22] 55% more productive you can't

[00:46:24] use it without a coder like it would be

[00:46:26] ridiculous because it's just giving you

[00:46:28] snippets of code and hints of code

[00:46:30] and it doesn't always work and you have to

[00:46:32] and then you have to modify it to make it work

[00:46:34] for exactly what you're doing etc

[00:46:36] but that means I don't have

[00:46:38] to write every line it'll spit out

[00:46:40] 20 or 30 or 40 lines of code just

[00:46:42] from my typing just like Word does

[00:46:44] you know I start to type and it starts to finish the sentence

[00:46:46] yeah that is exactly what co-pilot

[00:46:48] for coding is doing it literally

[00:46:50] finishes your code it's not exactly

[00:46:52] right all the time sometimes it doesn't

[00:46:54] compile but it's better than what I used

[00:46:56] to do which is go to GitHub

[00:46:58] or go to a variety of open source

[00:47:00] places and start to look

[00:47:02] for open source examples because you want to start

[00:47:04] with an open source example often today

[00:47:06] for this function now it's just giving you

[00:47:08] that example this is it's

[00:47:10] a 55% of that means in

[00:47:12] theory we might be able to double the

[00:47:14] productivity of coders using this

[00:47:16] and so it should be a given in every

[00:47:18] coding shop that of course you're using

[00:47:20] co-pilot do you still

[00:47:22] code not as well as

[00:47:24] I you know you know here's

[00:47:26] the problem right as you move up in management

[00:47:28] and you move into architecture and things like

[00:47:30] you lose the ability

[00:47:32] to do it syntactically correctly

[00:47:34] right so I could still

[00:47:36] sit there and write pseudo code and people go

[00:47:38] okay I know what you mean let me go fix it

[00:47:40] but to sit there

[00:47:42] and do it you have to know the syntax

[00:47:44] and you have to remember the syntax

[00:47:46] exactly where does that bracket

[00:47:48] go and exactly where the comma goes

[00:47:50] and so if you don't do that all the time

[00:47:52] you lose that do you think Asia is something to do with it

[00:47:54] like the last time I tried to code was about

[00:47:56] two years ago and I just was

[00:47:58] really unbelievably

[00:48:00] strikingly bad at it and I used to

[00:48:02] you know I've put in my

[00:48:04] 10,000 hours on coding and

[00:48:06] I did too I just can't

[00:48:08] I really just can't do it anymore

[00:48:10] I don't know if it's a great question

[00:48:12] whether age has to do it or we

[00:48:14] simply just graduate to higher level thinking

[00:48:16] right you get to again architecture

[00:48:18] and systems level and think

[00:48:20] about where the world you know and then you just

[00:48:22] that's just not a skill okay could you

[00:48:24] do long division you might

[00:48:26] not be really good at it but if you've struggled

[00:48:28] with it for a while you could do long division but you haven't

[00:48:30] done it since I don't know fifth grade

[00:48:32] right nor have I but

[00:48:34] we were taught it but we've kind of forgotten

[00:48:36] how to do some of that because the calculator came

[00:48:38] out so I don't know I think

[00:48:40] if you said I am going to learn Python

[00:48:42] and I'm going to take a month to learn it

[00:48:44] or two months at the end of

[00:48:46] two months you'd be every bit as good as you ever were

[00:48:48] and you might be more thoughtful about it you probably

[00:48:50] be really fast but you might

[00:48:52] not enjoy it you might say this is

[00:48:54] dumb because I could be the whole other level

[00:48:56] so I look I think copilot is

[00:48:58] a miracle for coders it's fascinating

[00:49:00] but no I wish I

[00:49:02] I wish I coded as fast as I did when I was at IBM

[00:49:04] and I could just rip code out thousands

[00:49:06] of lines you know yeah but

[00:49:08] those days are behind me

[00:49:10] and you know you got to give up something right you just can't do

[00:49:12] everything all the time I guess so yeah

[00:49:14] sometimes I try to do everything all the time but

[00:49:16] it doesn't work but

[00:49:18] Kevin Sarace you know thank you so much

[00:49:20] there's been so many questions

[00:49:22] and angst and anxiety

[00:49:24] and curiosities about

[00:49:26] AI just from everywhere and I've had

[00:49:28] my own frustrations just seeing

[00:49:30] how people have been abusing

[00:49:32] the words AI and what they do

[00:49:34] but look they're trying to make a living too so I don't blame them

[00:49:36] but thanks so much for

[00:49:38] helping clearing the air and all this stuff

[00:49:40] and good luck with all your entrepreneurial

[00:49:42] efforts particularly the ones I'm invested in so

[00:49:44] yes exactly exactly

[00:49:46] well we'll keep we'll keep flowing along

[00:49:48] and we'll get back together

[00:49:50] and maybe in six months and see where AI

[00:49:52] is then but I think I think all this hype

[00:49:54] is going to settle out and it's going to settle

[00:49:56] into like I said what Excel

[00:49:58] is today it's a great assistant it's a great

[00:50:00] tool for us it doesn't run itself

[00:50:02] normal AI

[00:50:04] you have to prompt it you have to get stuff back

[00:50:06] you have to edit it you have to be thoughtful about

[00:50:08] it but what a powerful tool

[00:50:10] and what a great time to live like I can't imagine

[00:50:12] living in another time

[00:50:14] I'm sure our parents thought oh there's color TV

[00:50:16] look at that okay this is way

[00:50:18] better than color TV right no

[00:50:20] and you know because of

[00:50:22] the increase in the speed of computers

[00:50:24] it's created

[00:50:26] exponentially growing industries in many

[00:50:28] other industries that

[00:50:30] required the computational power so

[00:50:32] initially computer technology was

[00:50:34] the only industry

[00:50:36] exponentially growing but now that it's

[00:50:38] exponentially grown it's given it's unlocked the power

[00:50:40] for other industries to

[00:50:42] that's what makes this super exciting time now

[00:50:44] so I agree it's amazing

[00:50:46] it's the best time to live yeah

[00:50:48] so thank you so much Kevin and I'll talk to you soon

[00:50:50] yeah a pleasure