Are AI Agents and Robots Taking Over Jobs—and Extending Our Lifespan?
What if AI agents could book your flights, manage your work, and even double the human lifespan? From jaw-dropping advancements in autonomous agents to robots sprinting at Olympic speeds, this week’s AI news is both thrilling and unnerving.
Transform your business with AI? Don’t miss the AI Business Transformation Course. Use the promo code LeveragingAI100 for $100 off the next cohort starting February 17th. https://multiplai.ai/ai-course/
In this episode of the Leveraging AI podcast, we dive into OpenAI’s latest release, "Operator," a groundbreaking tool that mimics human actions to complete tasks on your browser. We also explore Perplexity’s AI assistant, revolutionary breakthroughs in robotics, and predictions from leading AI thinkers about a world where humans and AI coexist in the workplace—and beyond.
In this session, you’ll discover:
- How OpenAI’s "Operator" redefines the role of personal and professional assistants.
- The growing impact of robotics on industries—and what it means for blue-collar jobs.
- How AI breakthroughs might enable humans to live beyond 150 years.
- What business leaders can learn from Salesforce’s CEO about adapting to an AI-driven workforce.
- Real-world AI applications transforming education, healthcare, and workplace efficiency.
- The challenges facing AI adoption, from regulation to trust, and how to navigate them.
If you found this episode valuable, please rate, review, and share it with others who want to stay ahead in the AI revolution. Don’t forget to connect with your host, Isar Meitis, on LinkedIn for ongoing insights into leveraging AI for growth and innovation.
About Leveraging AI
- The Ultimate AI Course for Business People: https://multiplai.ai/ai-course/
- YouTube Full Episodes: https://www.youtube.com/@Multiplai_AI/
- Connect with Isar Meitis: https://www.linkedin.com/in/isarmeitis/
- Join our Live Sessions, AI Hangouts and newsletter: https://services.multiplai.ai/events
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Hello, and welcome to a weekendnews episode of the Leveraging
(00:03):
AI podcast, the podcast thatshares practical, ethical ways
to leverage AI to improveefficiency, grow your business
and advance your career.
This is Isar Meitis, your host,and we had a completely
explosive week when it comes toAI news this week.
We are going to talk even moreabout AI agents, about
infrastructure for AI in the U.
S.
Thanks About crazy advancementsin robotics and about some
(00:28):
incredible scientificdiscoveries that potentially can
expand our lifespan by a lot.
And so lots to talk about.
We'll finish with a long list ofreally cool stuff, features,
capabilities, models that havebeen released this week.
But we have a few big topics todive into.
So let's get started.
(00:50):
the first big piece of news thatwill lead us into the discussion
about agents is that OpenAI justreleased Operator.
Operator is something theypromised and discussed late last
year, and they finally havereleased it.
What operator does is it takesover your browser to complete
tasks.
So you can give it tasks such asto book your flights or hotels
(01:11):
or order food or find your placein a restaurant or a lot of
other things it can in theory dothis on any website because what
it actually does is itunderstands the tasks you're
giving it.
It breaks it down on its own tosub tasks and steps that it
needs to complete.
And then it uses the keyboardand mouse to manipulate
everything on the browser,basically like a human does,
(01:33):
meaning in theory, it can doeverything that we can do as
humans with computers, which isMost of what we do to run the
world in our day to day and soon.
Now, the reality is right now,it's more focused at least as
far as open AI has promoted invideos that they've shared into
more personal life stuff.
Meaning, as I mentioned, bookingflights, ordering food, finding
(01:55):
recipes, and then ordering theingredients and so on.
It seems to be very promisingand the direction is very, very
clear, by the way, right nowit's only available to pro
users.
So the people who are paying 200a month, once this thing evolves
and it can really do things inour work.
beyond just day to day basicstuff, it will be worth a lot
more than 200 a month, becausein theory it can do everything
(02:18):
that we can do.
And a few things I anticipateone is that we'll get the
capability to train it onspecific tasks.
Basically, let me show you fivetimes, 10 times, 20 times how
I'm doing this.
You will learn from that andthen you will be able to.
Do this on its own, regardlessof what are the steps, which
software I'm using, how am Ibouncing back and forth between
different things?
And the other thing is that I'manticipating that we'll get the
(02:40):
ability to give it guardrails,meaning to define her software
per.
window per use case, what it canand cannot do so we can have
more control and make sure thatit doesn't do anything
catastrophic or just whateverdamages that it can create.
But as a first step, this isreally, really promising and
(03:00):
really, really scary.
Now, a few things I've alreadydone and done right is that it
requires your confirmationbecause before it takes any
final steps like booking yourflights or ordering food and so
on, it will require yourapproval or participation in
order to do things like enteringpasswords or credit card numbers
and so on.
So there are.
(03:21):
Already several differentguardrails put in place.
It's also not saving and nottraining on any of the
information it's capturing asscreenshots, which is how it
knows what's happening on thescreen.
So there are a few good steps inthe right direction as far as
safety and security.
But there's still a lot of openquestions on exactly what it can
and cannot do, how we limit itsaccess to things it shouldn't
(03:43):
have access to, and so on and soforth, including anti phishing
capabilities, because what if itfollows stuff that it thinks is
good and then it shares yourinformation in places it
shouldn't?
Like, there are a lot of openquestions that needs to be
addressed, but it's a very,strong step in the direction of
real autonomous agents that canreally act on our behalf and
(04:03):
without any technical skills.
So there's multiple platformsout there today that allow you
to develop multiple differenttypes of agents, but for all of
them, you need to have technicalskills to an extent.
Some of them more, some of themless either low code or even no
code, some of them, but youstill need technical skills.
This thing requires nothing.
You're literally just open it,ask it to do something and it
goes and does it.
(04:23):
And as I mentioned, I will bereally surprised if it doesn't
come with, train me how to dothis thing.
And I'll be able to do it foryou sometime in the next few
months.
In parallel, Perplexity launcheda new app that is an AI
assistant.
This app is currently availableon Android, and what it allows
Perplexity to do is to accessmultiple apps that you're doing
(04:43):
right now and help you do thetasks in these apps.
So examples could be, Getting anUber to pick you up from a
specific place at a specifictime and take you to a specific
location.
Just by saying it, I need anUber to take me to the airport
at 4 30 PM and it will doeverything it needs to do within
the app to make it happen.
Or find and play music from amovie that you don't even
(05:04):
remember who was the artist orwhat was the name of the song,
but you know, it was the themesong of a specific movie.
It will find the movie and theywill play it for you on Spotify.
Now, in addition to just apps,it's also fully multi modal,
meaning it can go and use yourcamera to see things in the real
world and then take actionsbased on that, such as looking
(05:24):
at a destination and giving youdirections on how to get there
or looking at a book and givingyou quotes or finding you what
else to read that is similar tothat book, and so on and so
forth.
Again, this is another step inthis particular case, integrated
into our phones, operating appsversus operating a web browser,
but the direction is clear.
Very different, by the way, fromOpenAI.
(05:45):
This functionality fromPerplexity is free right now.
So if you want to give it accessto do things on our behalf on
your phone and try it as a superassistant on your phone, you can
download it and use it rightnow.
Now.
To expand on that to what arethe implications of these agents
and ideas, there has been a veryinteresting panel this past week
(06:06):
in the World Economic Forum inDavos that included some high
figures such as Mark Benioff,the CEO of Salesforce, and Dario
Amodei, the CEO of Anthropic andmany other people.
And they've shared a lot ofreally interesting insights to
the future.
Nothing really new from whatthey said before, but it was
(06:27):
crystal clear from everythingthat we're talking about that
the future and even the nearfuture is going to be
dramatically different from whatwe know today.
So Mark Benioff, as an example,shared again that he thinks
we're going to see a dramaticfundamental shift in the
workforce composition from CEOsmanaging humans.
To CEO managing humans and AIagents with probably a growing
(06:49):
and growing percentage of AIagents.
He was talking about some of thethings that they're doing in
Salesforce.
So they are planning totransition support agents.
Two sales positions because AIsupport agents are doing such a
great job and they need lesspeople doing that.
And they can shift them tosales.
They're talking about that.
Several employee roles are beingredeployed as AI handles some of
(07:13):
these tasks.
So a few things to expand onthat.
One is.
That if a company is on a growthtrajectory, if you can grab more
market share, instead of lettingpeople go, you can use people
who understand the company,understand your product, and
then Sarah services and haveexperience in working in your
industry and just shift them tonew roles, which is great.
From the roles that AI can dofor them right now.
(07:35):
The other thing is on the moregranular level is that it's
going to be more and more a taskoriented world versus a
profession oriented world,meaning AI will be able to do
specific tasks out of specificroles, and then people's roles
will shift into other things asAI can do more and more of these
tasks.
And that's the lens throughwhich you need to start
(07:56):
evaluating your business,meaning look at the task level,
figure out what AI can do today,what are we able to do in the
next two years, and how you canget to the most effective
outcome, Preferably by shiftingemployees rather than letting
them go and allowing them to dostuff that will enable the
business to grow faster and moreefficiently.
Now in the same forum, as Imentioned, Dario Amadei, the CEO
(08:16):
of Anthropic was there as well.
And he made some crazypredictions.
One of them is that he believesthat AI will enable doubling
human lifespan within a decade.
So within the next 10 years,allowing people to live to 150,
160 years old, which I findvery, very extreme.
Now he went back to his claimthat he believes that AI system
(08:37):
will surpass most humanscognitive capabilities by 2026
to 2027.
He also mentioned that they'realso developing a virtual
collaborator AI agent forworkplace tasks.
So similar to what we justtalked about.
From both Salesforce and open AIand everybody else, agents that
we've got to run within theworkplace and learn and execute
(08:58):
tasks that will replace peopledoing the same tasks.
Now, the forum also discussedsome challenges of what may slow
down this incredibly fast AIrevolution.
One of it is physical worldlimitations, self driving cars
and needs to be built and thereneeds to be potentially some
additional infrastructure tomake them work more effectively.
(09:20):
Obviously, bureaucracy andregulation.
hurdles will slow AIimplementation down.
Public trust.
So I work with a company thatprovides call center solutions.
They.
As you may or may not know, youcan now have voice agents
replace human agents as callcenter agents and do most of the
work pretty well as we heardfrom the CEO of Salesforce.
(09:42):
And yet many companies, manyclients of my client are
refusing to let AI handle theircalls on their behalf because
they don't trust it yet.
So there's going to be thingsthat are going to slow down the
actual implementation.
But from a technologicalperspective, 2025 is going to be
a year where Of extreme rapidchange and as these agents
evolve and become more capableand safer to use, they're going
(10:06):
to have a profound implicationon literally everything we know
from our day to day, likeordering food or buying clothes
or looking for a restaurant ormaking travel plans all the way
to our day to day work in theworkforce.
In addition, I got to listen toa really interesting interview
this week with Sam Altman on theRethinking Podcast.
(10:26):
And I got to listen to SamAltman a lot because he does a
lot of interviews and I try tolisten to all of them to see how
he thinks and what he believes.
But this was a little differentbecause the guy interviewing him
is a psychologist who looks athuman nature and how does AI
reflect on that?
And so some of the questionswere different, but the bottom
line is very simple to whatDario is saying, where he thinks
(10:46):
that by 2026 or maybe 2027, AIsystem will surpass most humans
at almost everything.
He believes that while AI willbe highly capable and smarter
than all of us in the immediatefuture, it will have a
relatively gradual impact on ourday to day just because of the
way the world works.
So he is claiming that yes, weare close to AGI and that O3 is
(11:09):
incredible and that if somebodywould have told us three years
from now that we're going tohave a tool that powerful in
three years, What would we thinkthe impact is going to be on the
world?
And he thinks everybody wouldhave said, well, it will change
everything we know.
And the reality is, well, verylittle has changed.
Well, to an extent, I agree withhim because of some of the
reasons that I just mentionedthat slows AI adoption down.
But the reality is as peoplelearn and as groups,
(11:33):
governments, industries learnhow to use this, it will have an
impact.
A very, very profoundimplication on everything we
know, literally everything.
Now a few really interestingthings that Sam said that is
worth thinking about.
One he said that his child willlive in a world where AI is
smarter than humans, period.
(11:53):
And there's not going to be anyhumans in the future that will
ever be smarter than AI.
And he's saying That's not goingto be a problem because that's
just going to be the way it is.
Part of the conversation wasabout what's important for
humans in the future.
And he mentioned that the futurework will focus more on asking
the right questions rather thanfinding the right answers
because AI will be able to findthe answers for you.
(12:16):
He said that human will stillstay socially connected despite
AI superior capabilities becausethat's our nature is to stay
socially connected.
And we are going to appreciatehuman relationships even
further.
And I'm a hundred percentagreeing with that.
I teach that in my courses.
I'm a huge believer that humanrelationships will become even
more important, both on thepersonal lives and on our
(12:39):
productivity in business.
And then the last thing he saidabout properties people should
have in order to be moresuccessful, he said that.
raw intellectual horsepower willbecome less valuable than
adaptability, right?
Because AI will know stuff forus and being able to use it,
adapt quickly to changing toolsand so on will become
significantly more important.
(13:01):
Now, Sam sees AI as the mostdevelopment and interesting
scientific revolution of ourlifetime, partially because it's
going to enable a lot of otherstuff in other scientific
fields.
I mentioned before that DarioAmadei is thinking this can
double our lifespan.
Well, OpenAI, in collaborationwith Retro Biosciences, which is
a company that Sam personallyinvested in about 180 million,
(13:25):
announced earlier this monththat using GPT 4b, which is a
model that they developedspecifically to do protein
modification, they are now ableto to significantly improve a
process that is called Yamanakafactors.
And that process enables to takehuman skin cells and convert
them into stem cells.
(13:47):
Stem cells are the cells thatare the source of everything in
our bodies.
Basically, they can turn intoany other cell.
This is the first step ofbabies.
It starts with stem cells thatthen diverge into all the other
types of cells in our bodies.
So being able to take anexisting cell and convert it
into a stem cell could open acompletely new universe of
(14:09):
health care solutions for peopleand animals, and that process
was known before, but using thisnew a I model GPT for B, they
can do it 33 percent moreeffective than any human would
have done this so far Thisepisode is sponsored by the AI
Business Transformation course.
(14:29):
I have been personally teachingthe AI Business Transformation
course since April of 2023.
I've been teaching the course atleast once a month in many
months.
I've done two Courses a month,but most of these courses are
private, meaning companies,organizations hire me to train
their people.
And about once a quarter, I havethe bandwidth to actually open a
course to the public, which Ireally love doing because it
(14:51):
enables you literally every oneof you to come and learn with me
with an amazing cohort of peoplelike you who are business people
who are looking to learn how toimplement AI successfully in
your business.
We just opened the registrationfor the next cohort, which is
going to start on February 17th,which is a Monday at noon
Eastern.
It's two hours a week for fourweeks.
So eight hours with me, plusoffice hours every Friday to
(15:13):
come and ask me anything youwant about the progress or about
anything else in AI.
And because you're listeners ofLeveraging AI, you can use the
promo code LeveragingAI100 inorder to get 100 off the course.
Don't miss this opportunity.
As I mentioned, we open thecourse to the public about once
a quarter, meaning if you wantto gain the maximum benefits you
can from AI in 2025, and youshould come join us in February,
(15:38):
because most likely the nextcourse is going to be in the May
timeframe.
I really hope to see you there.
Now, back to the show, stayingon the topic of discoveries.
Demis Hassabis just announcedthat AI designed drug trials are
projected to start in 2025.
So, isomorphic Labs is one ofAlphabet's companies, so a
(15:59):
sister company to Google, andthey are working on drug
discovery.
They have partnerships with someof the largest drug companies in
the world, and they're going tobegin AI designed drug trials by
the end of this year, by the endof 2025, and they're targeting
of issues like oncology, andcardiovascular, and neuro
(16:21):
problems and diseases, and theyare claiming that they can
shorten the discovery process by10X, meaning instead of five to
10 years, it could happen in afew months.
These are all very promisingideas that can help us fight
some of the biggest diseasesthat we have in the world today
within the next few years, whichmay relate to the crazy
(16:43):
prediction by Dario Amadei ofdoubling our lifespan.
Now, If that's not enough, Axioshas reported, presumably OpenAI
are preparing to announce abreakthrough in agent
technology, and they will deployor at least have what they call
super agents that are PhD levelon almost every topic.
(17:03):
Now, they're claiming thatmultiple sources indicate AI
companies are exceeding theirinternal development projections
and that they're achieving AGIand beyond within this year that
obviously caught like wildfireon X and rumors started flying
left and right, startedconnecting it to the meeting
that Sam Altman has behindclosed doors on January 30th
(17:26):
with government officials.
the rumors basically went out ofcontrol.
So Sam Altman went to Axe andshared the following tweet.
He said, Twitter hype is out ofcontrol again.
We are not going to deploy AGInext month, nor have we built
it.
He also shared that the companyhas some cool stuff coming.
We already know some of it liketasks in operator, maybe there's
(17:47):
more coming, but he warned AIfans needs to cut their
expectations by a hundred X.
So contradicting ideas andconcepts coming from multiple
directions.
But what's very obvious is thatagents are here.
They're going to become more andmore capable and more and more
available to anyone to developand use.
(18:09):
And these agents will be able todo stuff that we can do right
now.
And that will, on some cases,Remove jobs on some cases,
enhance jobs, enhance happinessof employees, because you'll be
able to do things that are morevaluable and it can impact some
really big questions likescientific discoveries and
accelerate them dramatically.
Now, if you remember last week,we talked a lot about
(18:30):
infrastructure as well.
So that's our next topic today.
Again, President Trump, that isnow not president elect, but the
actual president, announced a500 billion infrastructure
project that is started by threetech giants.
The project is called Stargateand it involves OpenAI,
SoftBank, and Oracle.
And the plan is to initiallydeploy 100 billion of
(18:53):
investment, scaling to half atrillion dollars, all in AI
infrastructure on U.
S.
soil.
Now, the project aims to build20 data centers in the U.
S., at least 500, 000 squarefoot The first facility is
already under construction inTexas, and it perfectly aligns
(19:13):
with OpenAI's roadmap that weshared with you last week.
And their estimate that there's175 billion in global funds that
are currently awaiting AIproject investments that OpenAI
is trying to bring into the U.
S.
Now, while this is on one handpromising and interesting and
shows that there's a reallyserious investment and that the
new administration is fullysupportive of this, there are a
(19:36):
lot of open questions such aswhere is the power going to come
from?
Where is the coolingcapabilities going to come from?
But what the currentadministration is saying is a
keeping the U.
S.
ahead of the competition, mostlyChina at this point, and B,
generating new jobs, potentiallyreplacing the jobs that would be
lost by AI.
So this particular project ofinfrastructure is expected to
(19:56):
generate a hundred thousandjobs.
Now that sounds like a reallybig number, but if you think
about the fact that AI agentsmight be around the corner and
they'll be able to do, well,more or less everything that we
can do in front of a computer,and then robots can do more or
less the rest, and that's ournext topic.
100, 000 jobs is not that many,but at least it's creating some
jobs.
Now, going back to the positiveside of things, the World Bank
(20:19):
just did a very interestingstudy about AI tutoring in
Nigeria.
And what they've learned that byintegrating AI personal tutoring
to every single student in theway they like to learn where
they can ask questions and getassistance in their own pace has
made a dramatic improvement inthe results of all students
taking it.
And it basically took theequivalent of two years.
(20:42):
of learning and was able toachieve the same results in six
weeks.
The students who participated inthis test outperform 80 percent
of other educational solutionsin developing nations, and it
showed benefits topics acrossEnglish, AI knowledge, digital
skills, and so on.
It even had an improvementperformance in other curriculum
(21:04):
subjects outside of the program,just because it gets the
students to be excited aboutlearning and being able to be
more successful.
I said that many times before, Ithink the education system in
the 20th century has made verylittle progress in the last
hundred years.
Most of the things that aretaught are still taught by a
teacher with a board behind himin front of a class with a lot
(21:25):
of people that does not make anysense in the year of 2025.
And while AI represents a verybig risk to that, it represents
the biggest opportunity we hadto really unlock fast learning
that is tailored to the needs ofthe individual, both in the pace
that they need to learn thethings they get stuck on and the
way they like to learn, whetherit's playing games, listening to
(21:45):
stories, reading books, watchingvideos, whatever works for them,
or any combination that will betailored to the specific
individual needs and allowingthem to achieve significantly
better results faster.
And now there's a first researchactually showing it again, two
years of learning progresswithin six weeks of using an AI
tutor, absolutely mind blowing.
And really exciting from myperspective.
Now I told you, we also got totalk about blue collar jobs in
(22:07):
this episode.
Well, a few interesting thingshave happened this week as far
as releases.
So a Chinese company hasreleased a new version of the
robot called black Panther.
So it's black Panther versiontwo, and it can sprint a hundred
meters in under 10 seconds.
So.
As the name suggests, it lookslike a panther, like it looks
like a four legged animal, butit can run crazy fast, meaning
(22:30):
it can run faster than mostpeople on the planet.
This is an Olympic record kindof speed.
The abilities of these robotsare accelerating very, very
fast.
Take a robot like this tohelping in any kind of natural
disasters where they can run onuneven surfaces and get to
places very, very fast.
It's actually relatively small,so it's about two feet high, so
it can crawl under and getaround things relatively easy
(22:53):
and think like, again, a bigpat.
And so.
That's very promising.
There's always the flip side.
This can be used for militarypurposes, and then it can be in
the battlefield and do a lot ofdamage if it's not controlled
properly.
So there's always the good andthe bad with each and every one
of these deliveries.
Unitry G1 is a robot that wetalked about many times before.
Unitree is one of the Chinesecompanies that are developing
the most advanced robots thatthere are out there today.
(23:15):
And they have a big robot calledH1, which was the first thing
that they developed, but theirfocus recently have been on G1,
which is its little brother.
It's about four feet tall andthe price point for it right
now, they're already pre sellingit.
So if you want one, you can goand order one, is 16, 000 for
the base model.
That's Very reasonable for thestuff that it can do.
And what they just demoed isthat they demoed the latest
(23:36):
version of G1 actually runningand jogging across a huge
variety of terrains.
So first of all, it can moveagain, a lot faster than a
previous robot, but it can do iton site slope on very steep up
and down slopes on uneventerrains with rocks and debris
and so on.
And so these robots are becominghighly, highly, highly, highly,
capable and they are becomingreally cheap.
(23:57):
Now, not yet, but we talkedabout this last week, that Elon
Musk is claiming that Tesla isgoing to manufacture hundreds of
thousands of these robots in2026.
So I hear from now inpotentially millions in 2027.
And so the mass production ofthese is just around the corner.
Relating to that is an economistnamed Rubini, who got known as
(24:18):
Dr.
Doom for his gloomy economicforecast is warning about the
rapid development of theserobots their potential impact on
the blue collar job market.
He is.
Estimating that the market ofthese robots will reach 7
trillion by 2050 and that theserobots will start disrupting the
labor market in the next year ortwo.
(24:40):
So again, aligned with thetimelines that Elon Musk was
sharing in his predictions onhow many of these are going to
manufacture.
So if you put in all of thistogether, the capabilities of AI
agents combined with the ofrobots.
And you see a huge disruption toalmost every job out there.
And the pace in which it'spotentially happening.
And yes, we talked about thingsthat will slow it down, but the
(25:02):
pace in which it is potentiallyhappening is a disruption nobody
is ready for.
Now let's dive into some rapidfire announcement on some new
features, capabilities, andother announcement that happened
this week.
First of all, Anthropicannounced voice chat memory
features for Claude.
So the features that are comingand haven't been released yet is
two way voice conversation,similar to what gemini and
(25:23):
Chachapiti already have memoryof past interactions.
Again, something that alsoChachapiti and Gemini already
has and personalized chatcapabilities, which also open AI
already has.
So all these three capabilitiesare coming to Claude.
I must admit I use Claude a lotand I like it for many different
reasons.
One of the things that drives mecrazy is that it still doesn't
(25:43):
have internet access.
And yes, I know you can get itto connect to the internet
through third party tools.
But I don't understand why.
Anthropic hasn't released thatcapability yet.
So if somebody from Anthropic islistening, please add that
functionality, to your toolbecause I love using it, but
it's just really far behind theother tools from that particular
perspective.
Now, another really interestingthing that Anthropic just
(26:04):
launched and that is availableis feature called citations and
citations allowed to ground AIresponses in source documents,
dramatically reducinghallucinations, and it also
provides references to specificsentences and passages in the
original data.
Now it's going to be availablethrough the API to Cloud 3.
5 SONNET and Cloud 3.
(26:24):
5 HYCU and it's going to beavailable through their API or
through Google Vertex where youcan also get access to the
Anthropic platform.
Just as a quick note, you can dostuff like that today In any
chat tool that you're using, Ido this every time I reference
information, every time I ask itquestions about information that
I provided, I ask it for exactquotes, and I ask it for the
(26:45):
name of the document, the page,And the specific paragraph where
they found the quote so I can goand verify the information that
it's giving me.
And that dramatically reduceshallucinations and also allows
me to verify the informationvery, very quickly.
But having it available nativelyin the API would be absolutely
awesome.
Now, similar to this, Perplexityreleased what they call Solar
Pro API, which does similarthings.
(27:06):
It allows you to get citationsand customize sources in the API
to answers that it's giving you.
While focusing on speed andaffordability, that's
perplexity.
So again, the concept ofhallucinations is something
that's going to be reduceddramatically this year as well,
which will give us much bettertrust on these tools right now,
because right now it's a hit ormiss.
(27:27):
And yes, the hit or miss rightnow is probably 80 percent
correct and 20 percentincorrect, but there are many
use cases where that is just notacceptable.
Now staying on the topic of newmodels and capabilities, we
spoke A lot last week about deepseek version three, which came
almost out of nowhere and tooknumber seven on the leaderboard
On the LM sys chatbot arena.
Well, they just released athinking model that integrates
(27:50):
actually the ability to do testtime compute.
So think as it's doing the taskin combination with deep
research capabilities and thatmodel jumped into number four of
the chapel arena, actuallytaking number one on several
different aspects.
I must remind you that thismodel was developed on tens of
cents on the dollar comparedwith the amount of money that
(28:10):
was invested in the leadingmodels from open AI, Anthropic,
Gemini, and so on.
And because of that, they canprovide this kind of capability
for extremely low cost.
Now you can use it through theAPI, you can use it on their
chat, and you can also take anduse it however you want because
it's a full open source modelthat you can actually go and
(28:30):
grab from their website or fromHugging Face, install it on your
servers or on your computer andrun it locally with your data
without sending it anywhereelse.
This is to me is a huge promisefor a future in which we may not
need 500 billion of investmentin data centers, and then the
cooling and power and theenvironmental damage that comes
(28:51):
with it.
Because if they can achievethese kinds of results with
significantly less compute andsignificantly less time, there
is hope that other companies andhopefully will do the same.
And if competition tells usanything, that's what's going to
happen.
And again, that's why I findthis to be very, very good news,
despite the fact that this comesfrom China and not from the U S.
Going back to another piece ofaddition about agentic news,
(29:14):
Microsoft just released AutoGen0.
4 and that provides thecapability to users of the
Microsoft ecosystem to developand integrate agents directly
ecosystem offering custom toolsto build and deploy agents
across enterprises.
Some interesting things aboutthis model is that they're
achieving cost effectiveefficiencies that were not
(29:36):
possible before, while enablingasynchronous processing, meaning
multiple agents can run inparallel, one can do data
collection, one can do parsing,one can do reporting, one can
do, communication and so on, allwhile being orchestrated by a
centralized agent that willbasically be the project
manager.
Another interesting tool thatwas, announced this week is
(29:57):
Mistral, which is a French opensource language model company,
has came out with what theycalled CodeStral 2501, which is
now the most accurate codegenerator out there.
They're achieving 95.
3 FIM accuracy score inprogramming language, surpassing
OpenAI's API by over a millionby 2.
(30:17):
6%, which was the leader so far.
So GPT 01, and it has a contextwindow of 256, 000 tokens, which
allows you to write or review alot of code without model.
I'm reminding you that Mistralis an open source company.
So again, you can go and eitheruse it on their platform with
their API, or just get the opensource and run it locally for
yourself and now develop codebetter than anything else.
(30:41):
It's very, very obvious fromthese tools, as well as
announcements from people likeZuckerberg, that the world of
writing code is going to changeis not going to change, but it's
already changing dramatically.
If you remember, I told you lastweek that Zuckerberg said that
in 2025 coding agents will beable to do the work of mid level
software engineers, and thatthey're already doing it at
(31:03):
Meta.
And two really big pieces ofnews that I kept for the end
because we had too many bigitems in the beginning is iT
spending is projected to hit 5.
5 trillion dollars in 2025.
That's almost a 10 percentincrease for 2024, which was a
record breaking year.
And it's driven mostly by AIhardware needs.
(31:26):
Now Gartner just did a researchand they released it this week.
And they're saying That thegenerative AI is entering, and
I'm quoting the trough ofdelusion meant, which basically
means that there's going to bemore and more expenses and less
and less ROI, Which is astandard phase in the cycle of
hypes and the main concern isthat AI sectors that are showing
(31:48):
insane growth.
but are not really showing bigdifferentiation between every
one of them is raising questionsabout the sustainable value
creation that they can actuallykeep on delivering.
Again, time will tell what isgoing to win, whether it's going
to be this amazing technology orsociety slowing it down.
But it's very obvious to me thatthis is moving forward.
(32:09):
Despite these concerns, doesn'tmean anything about valuations
because literally every singleround that happened to companies
recently has at least doubledthe valuation of companies.
And the previous valuation wasSix to 12 months ago.
So companies valuations aredoubling every six to 12 months,
and that's obviously notsustainable.
Whether the hype is justified,we'll have to wait and see from
(32:31):
everything we're seeing rightnow.
I would say yes, it's too earlyto tell what's gonna be the
actual ROI on that.
And the other piece that isreally, really interesting is
that Google published a researchpaper about what they call the
Titan architecture and It ispotentially the first
significant major advancementssince the transformer paper back
(32:52):
in 2017.
So those of you who don't knowthe story, the paper called
attention is all you need.
That was the paper that Googleresearch have released in 2017
is basically the baseline foreverything AI that we know
today.
And they now released a newpaper called the Titan, which
builds on top of transformer.
So it's not totally replacingthem, but the idea here is
building a neural long termmemory alongside the short term
(33:16):
memory and basing the long termmemory on what they're calling
surprise based learning, whichthey claim is how human
cognitive patterns work.
So basically it's going to lookfor stuff that is out of the
ordinary, that is not alignedwith its existing patterns in
order to consider it new.
And then learn that as a newthing, early benchmarks already
showing superior results toexisting Language models across
(33:39):
several different use cases.
So it'll be very interesting tosee if this really drives the
next cycle of innovation in theactual architecture of how all
these AI models work.
We are going to be back onTuesday with another fascinating
how to episode with a expertsharing with you a specific AI
use case that you can startimplementing in your business
(34:00):
immediately.
If you have not yet ranked andrated this podcast, I would
really appreciate it.
It really helps me get to morepeople and it allows you to play
an important role in deliveringAI education for everyone.
And so please open your appright now, share this podcast
with as many people as you thinkcan benefit from it, and give us
(34:21):
whatever rating and review youthink we deserve.
Hopefully I've earned a fivestar review from you, but
whatever you think, write itdown there, give me feedback.
I actually read all thesecomments.
I would love to hear what youthink.
You can also connect with me onLinkedIn and let me know what
you think about the podcast.
I love chatting with you aboutthe podcast and on things that
you think can make it better anduntil next time, have an amazing
(34:43):
weekend.
Hello, and welcome to a weekendnews episode of the Leveraging
(00:03):
AI podcast, the podcast thatshares practical, ethical ways
to leverage AI to improveefficiency, grow your business
and advance your career.
This is Isar Meitis, your host,and we had a completely
explosive week when it comes toAI news this week.
We are going to talk even moreabout AI agents, about
infrastructure for AI in the U.
S.
Thanks About crazy advancementsin robotics and about some
(00:28):
incredible scientificdiscoveries that potentially can
expand our lifespan by a lot.
And so lots to talk about.
We'll finish with a long list ofreally cool stuff, features,
capabilities, models that havebeen released this week.
But we have a few big topics todive into.
So let's get started.
(00:50):
the first big piece of news thatwill lead us into the discussion
about agents is that OpenAI justreleased Operator.
Operator is something theypromised and discussed late last
year, and they finally havereleased it.
What operator does is it takesover your browser to complete
tasks.
So you can give it tasks such asto book your flights or hotels
(01:11):
or order food or find your placein a restaurant or a lot of
other things it can in theory dothis on any website because what
it actually does is itunderstands the tasks you're
giving it.
It breaks it down on its own tosub tasks and steps that it
needs to complete.
And then it uses the keyboardand mouse to manipulate
everything on the browser,basically like a human does,
(01:33):
meaning in theory, it can doeverything that we can do as
humans with computers, which isMost of what we do to run the
world in our day to day and soon.
Now, the reality is right now,it's more focused at least as
far as open AI has promoted invideos that they've shared into
more personal life stuff.
Meaning, as I mentioned, bookingflights, ordering food, finding
(01:55):
recipes, and then ordering theingredients and so on.
It seems to be very promisingand the direction is very, very
clear, by the way, right nowit's only available to pro
users.
So the people who are paying 200a month, once this thing evolves
and it can really do things inour work.
beyond just day to day basicstuff, it will be worth a lot
more than 200 a month, becausein theory it can do everything
(02:18):
that we can do.
And a few things I anticipateone is that we'll get the
capability to train it onspecific tasks.
Basically, let me show you fivetimes, 10 times, 20 times how
I'm doing this.
You will learn from that andthen you will be able to.
Do this on its own, regardlessof what are the steps, which
software I'm using, how am Ibouncing back and forth between
different things?
And the other thing is that I'manticipating that we'll get the
(02:40):
ability to give it guardrails,meaning to define her software
per.
window per use case, what it canand cannot do so we can have
more control and make sure thatit doesn't do anything
catastrophic or just whateverdamages that it can create.
But as a first step, this isreally, really promising and
(03:00):
really, really scary.
Now, a few things I've alreadydone and done right is that it
requires your confirmationbecause before it takes any
final steps like booking yourflights or ordering food and so
on, it will require yourapproval or participation in
order to do things like enteringpasswords or credit card numbers
and so on.
So there are.
(03:21):
Already several differentguardrails put in place.
It's also not saving and nottraining on any of the
information it's capturing asscreenshots, which is how it
knows what's happening on thescreen.
So there are a few good steps inthe right direction as far as
safety and security.
But there's still a lot of openquestions on exactly what it can
and cannot do, how we limit itsaccess to things it shouldn't
(03:43):
have access to, and so on and soforth, including anti phishing
capabilities, because what if itfollows stuff that it thinks is
good and then it shares yourinformation in places it
shouldn't?
Like, there are a lot of openquestions that needs to be
addressed, but it's a very,strong step in the direction of
real autonomous agents that canreally act on our behalf and
(04:03):
without any technical skills.
So there's multiple platformsout there today that allow you
to develop multiple differenttypes of agents, but for all of
them, you need to have technicalskills to an extent.
Some of them more, some of themless either low code or even no
code, some of them, but youstill need technical skills.
This thing requires nothing.
You're literally just open it,ask it to do something and it
goes and does it.
(04:23):
And as I mentioned, I will bereally surprised if it doesn't
come with, train me how to dothis thing.
And I'll be able to do it foryou sometime in the next few
months.
In parallel, Perplexity launcheda new app that is an AI
assistant.
This app is currently availableon Android, and what it allows
Perplexity to do is to accessmultiple apps that you're doing
(04:43):
right now and help you do thetasks in these apps.
So examples could be, Getting anUber to pick you up from a
specific place at a specifictime and take you to a specific
location.
Just by saying it, I need anUber to take me to the airport
at 4 30 PM and it will doeverything it needs to do within
the app to make it happen.
Or find and play music from amovie that you don't even
(05:04):
remember who was the artist orwhat was the name of the song,
but you know, it was the themesong of a specific movie.
It will find the movie and theywill play it for you on Spotify.
Now, in addition to just apps,it's also fully multi modal,
meaning it can go and use yourcamera to see things in the real
world and then take actionsbased on that, such as looking
(05:24):
at a destination and giving youdirections on how to get there
or looking at a book and givingyou quotes or finding you what
else to read that is similar tothat book, and so on and so
forth.
Again, this is another step inthis particular case, integrated
into our phones, operating appsversus operating a web browser,
but the direction is clear.
Very different, by the way, fromOpenAI.
(05:45):
This functionality fromPerplexity is free right now.
So if you want to give it accessto do things on our behalf on
your phone and try it as a superassistant on your phone, you can
download it and use it rightnow.
Now.
To expand on that to what arethe implications of these agents
and ideas, there has been a veryinteresting panel this past week
(06:06):
in the World Economic Forum inDavos that included some high
figures such as Mark Benioff,the CEO of Salesforce, and Dario
Amodei, the CEO of Anthropic andmany other people.
And they've shared a lot ofreally interesting insights to
the future.
Nothing really new from whatthey said before, but it was
(06:27):
crystal clear from everythingthat we're talking about that
the future and even the nearfuture is going to be
dramatically different from whatwe know today.
So Mark Benioff, as an example,shared again that he thinks
we're going to see a dramaticfundamental shift in the
workforce composition from CEOsmanaging humans.
To CEO managing humans and AIagents with probably a growing
(06:49):
and growing percentage of AIagents.
He was talking about some of thethings that they're doing in
Salesforce.
So they are planning totransition support agents.
Two sales positions because AIsupport agents are doing such a
great job and they need lesspeople doing that.
And they can shift them tosales.
They're talking about that.
Several employee roles are beingredeployed as AI handles some of
(07:13):
these tasks.
So a few things to expand onthat.
One is.
That if a company is on a growthtrajectory, if you can grab more
market share, instead of lettingpeople go, you can use people
who understand the company,understand your product, and
then Sarah services and haveexperience in working in your
industry and just shift them tonew roles, which is great.
From the roles that AI can dofor them right now.
(07:35):
The other thing is on the moregranular level is that it's
going to be more and more a taskoriented world versus a
profession oriented world,meaning AI will be able to do
specific tasks out of specificroles, and then people's roles
will shift into other things asAI can do more and more of these
tasks.
And that's the lens throughwhich you need to start
(07:56):
evaluating your business,meaning look at the task level,
figure out what AI can do today,what are we able to do in the
next two years, and how you canget to the most effective
outcome, Preferably by shiftingemployees rather than letting
them go and allowing them to dostuff that will enable the
business to grow faster and moreefficiently.
Now in the same forum, as Imentioned, Dario Amadei, the CEO
(08:16):
of Anthropic was there as well.
And he made some crazypredictions.
One of them is that he believesthat AI will enable doubling
human lifespan within a decade.
So within the next 10 years,allowing people to live to 150,
160 years old, which I findvery, very extreme.
Now he went back to his claimthat he believes that AI system
(08:37):
will surpass most humanscognitive capabilities by 2026
to 2027.
He also mentioned that they'realso developing a virtual
collaborator AI agent forworkplace tasks.
So similar to what we justtalked about.
From both Salesforce and open AIand everybody else, agents that
we've got to run within theworkplace and learn and execute
(08:58):
tasks that will replace peopledoing the same tasks.
Now, the forum also discussedsome challenges of what may slow
down this incredibly fast AIrevolution.
One of it is physical worldlimitations, self driving cars
and needs to be built and thereneeds to be potentially some
additional infrastructure tomake them work more effectively.
(09:20):
Obviously, bureaucracy andregulation.
hurdles will slow AIimplementation down.
Public trust.
So I work with a company thatprovides call center solutions.
They.
As you may or may not know, youcan now have voice agents
replace human agents as callcenter agents and do most of the
work pretty well as we heardfrom the CEO of Salesforce.
(09:42):
And yet many companies, manyclients of my client are
refusing to let AI handle theircalls on their behalf because
they don't trust it yet.
So there's going to be thingsthat are going to slow down the
actual implementation.
But from a technologicalperspective, 2025 is going to be
a year where Of extreme rapidchange and as these agents
evolve and become more capableand safer to use, they're going
(10:06):
to have a profound implicationon literally everything we know
from our day to day, likeordering food or buying clothes
or looking for a restaurant ormaking travel plans all the way
to our day to day work in theworkforce.
In addition, I got to listen toa really interesting interview
this week with Sam Altman on theRethinking Podcast.
(10:26):
And I got to listen to SamAltman a lot because he does a
lot of interviews and I try tolisten to all of them to see how
he thinks and what he believes.
But this was a little differentbecause the guy interviewing him
is a psychologist who looks athuman nature and how does AI
reflect on that?
And so some of the questionswere different, but the bottom
line is very simple to whatDario is saying, where he thinks
(10:46):
that by 2026 or maybe 2027, AIsystem will surpass most humans
at almost everything.
He believes that while AI willbe highly capable and smarter
than all of us in the immediatefuture, it will have a
relatively gradual impact on ourday to day just because of the
way the world works.
So he is claiming that yes, weare close to AGI and that O3 is
(11:09):
incredible and that if somebodywould have told us three years
from now that we're going tohave a tool that powerful in
three years, What would we thinkthe impact is going to be on the
world?
And he thinks everybody wouldhave said, well, it will change
everything we know.
And the reality is, well, verylittle has changed.
Well, to an extent, I agree withhim because of some of the
reasons that I just mentionedthat slows AI adoption down.
But the reality is as peoplelearn and as groups,
(11:33):
governments, industries learnhow to use this, it will have an
impact.
A very, very profoundimplication on everything we
know, literally everything.
Now a few really interestingthings that Sam said that is
worth thinking about.
One he said that his child willlive in a world where AI is
smarter than humans, period.
(11:53):
And there's not going to be anyhumans in the future that will
ever be smarter than AI.
And he's saying That's not goingto be a problem because that's
just going to be the way it is.
Part of the conversation wasabout what's important for
humans in the future.
And he mentioned that the futurework will focus more on asking
the right questions rather thanfinding the right answers
because AI will be able to findthe answers for you.
(12:16):
He said that human will stillstay socially connected despite
AI superior capabilities becausethat's our nature is to stay
socially connected.
And we are going to appreciatehuman relationships even
further.
And I'm a hundred percentagreeing with that.
I teach that in my courses.
I'm a huge believer that humanrelationships will become even
more important, both on thepersonal lives and on our
(12:39):
productivity in business.
And then the last thing he saidabout properties people should
have in order to be moresuccessful, he said that.
raw intellectual horsepower willbecome less valuable than
adaptability, right?
Because AI will know stuff forus and being able to use it,
adapt quickly to changing toolsand so on will become
significantly more important.
(13:01):
Now, Sam sees AI as the mostdevelopment and interesting
scientific revolution of ourlifetime, partially because it's
going to enable a lot of otherstuff in other scientific
fields.
I mentioned before that DarioAmadei is thinking this can
double our lifespan.
Well, OpenAI, in collaborationwith Retro Biosciences, which is
a company that Sam personallyinvested in about 180 million,
(13:25):
announced earlier this monththat using GPT 4b, which is a
model that they developedspecifically to do protein
modification, they are now ableto to significantly improve a
process that is called Yamanakafactors.
And that process enables to takehuman skin cells and convert
them into stem cells.
(13:47):
Stem cells are the cells thatare the source of everything in
our bodies.
Basically, they can turn intoany other cell.
This is the first step ofbabies.
It starts with stem cells thatthen diverge into all the other
types of cells in our bodies.
So being able to take anexisting cell and convert it
into a stem cell could open acompletely new universe of
(14:09):
health care solutions for peopleand animals, and that process
was known before, but using thisnew a I model GPT for B, they
can do it 33 percent moreeffective than any human would
have done this so far Thisepisode is sponsored by the AI
Business Transformation course.
(14:29):
I have been personally teachingthe AI Business Transformation
course since April of 2023.
I've been teaching the course atleast once a month in many
months.
I've done two Courses a month,but most of these courses are
private, meaning companies,organizations hire me to train
their people.
And about once a quarter, I havethe bandwidth to actually open a
course to the public, which Ireally love doing because it
(14:51):
enables you literally every oneof you to come and learn with me
with an amazing cohort of peoplelike you who are business people
who are looking to learn how toimplement AI successfully in
your business.
We just opened the registrationfor the next cohort, which is
going to start on February 17th,which is a Monday at noon
Eastern.
It's two hours a week for fourweeks.
So eight hours with me, plusoffice hours every Friday to
(15:13):
come and ask me anything youwant about the progress or about
anything else in AI.
And because you're listeners ofLeveraging AI, you can use the
promo code LeveragingAI100 inorder to get 100 off the course.
Don't miss this opportunity.
As I mentioned, we open thecourse to the public about once
a quarter, meaning if you wantto gain the maximum benefits you
can from AI in 2025, and youshould come join us in February,
(15:38):
because most likely the nextcourse is going to be in the May
timeframe.
I really hope to see you there.
Now, back to the show, stayingon the topic of discoveries.
Demis Hassabis just announcedthat AI designed drug trials are
projected to start in 2025.
So, isomorphic Labs is one ofAlphabet's companies, so a
(15:59):
sister company to Google, andthey are working on drug
discovery.
They have partnerships with someof the largest drug companies in
the world, and they're going tobegin AI designed drug trials by
the end of this year, by the endof 2025, and they're targeting
of issues like oncology, andcardiovascular, and neuro
(16:21):
problems and diseases, and theyare claiming that they can
shorten the discovery process by10X, meaning instead of five to
10 years, it could happen in afew months.
These are all very promisingideas that can help us fight
some of the biggest diseasesthat we have in the world today
within the next few years, whichmay relate to the crazy
(16:43):
prediction by Dario Amadei ofdoubling our lifespan.
Now, If that's not enough, Axioshas reported, presumably OpenAI
are preparing to announce abreakthrough in agent
technology, and they will deployor at least have what they call
super agents that are PhD levelon almost every topic.
(17:03):
Now, they're claiming thatmultiple sources indicate AI
companies are exceeding theirinternal development projections
and that they're achieving AGIand beyond within this year that
obviously caught like wildfireon X and rumors started flying
left and right, startedconnecting it to the meeting
that Sam Altman has behindclosed doors on January 30th
(17:26):
with government officials.
the rumors basically went out ofcontrol.
So Sam Altman went to Axe andshared the following tweet.
He said, Twitter hype is out ofcontrol again.
We are not going to deploy AGInext month, nor have we built
it.
He also shared that the companyhas some cool stuff coming.
We already know some of it liketasks in operator, maybe there's
(17:47):
more coming, but he warned AIfans needs to cut their
expectations by a hundred X.
So contradicting ideas andconcepts coming from multiple
directions.
But what's very obvious is thatagents are here.
They're going to become more andmore capable and more and more
available to anyone to developand use.
(18:09):
And these agents will be able todo stuff that we can do right
now.
And that will, on some cases,Remove jobs on some cases,
enhance jobs, enhance happinessof employees, because you'll be
able to do things that are morevaluable and it can impact some
really big questions likescientific discoveries and
accelerate them dramatically.
Now, if you remember last week,we talked a lot about
(18:30):
infrastructure as well.
So that's our next topic today.
Again, President Trump, that isnow not president elect, but the
actual president, announced a500 billion infrastructure
project that is started by threetech giants.
The project is called Stargateand it involves OpenAI,
SoftBank, and Oracle.
And the plan is to initiallydeploy 100 billion of
(18:53):
investment, scaling to half atrillion dollars, all in AI
infrastructure on U.
S.
soil.
Now, the project aims to build20 data centers in the U.
S., at least 500, 000 squarefoot The first facility is
already under construction inTexas, and it perfectly aligns
(19:13):
with OpenAI's roadmap that weshared with you last week.
And their estimate that there's175 billion in global funds that
are currently awaiting AIproject investments that OpenAI
is trying to bring into the U.
S.
Now, while this is on one handpromising and interesting and
shows that there's a reallyserious investment and that the
new administration is fullysupportive of this, there are a
(19:36):
lot of open questions such aswhere is the power going to come
from?
Where is the coolingcapabilities going to come from?
But what the currentadministration is saying is a
keeping the U.
S.
ahead of the competition, mostlyChina at this point, and B,
generating new jobs, potentiallyreplacing the jobs that would be
lost by AI.
So this particular project ofinfrastructure is expected to
(19:56):
generate a hundred thousandjobs.
Now that sounds like a reallybig number, but if you think
about the fact that AI agentsmight be around the corner and
they'll be able to do, well,more or less everything that we
can do in front of a computer,and then robots can do more or
less the rest, and that's ournext topic.
100, 000 jobs is not that many,but at least it's creating some
jobs.
Now, going back to the positiveside of things, the World Bank
(20:19):
just did a very interestingstudy about AI tutoring in
Nigeria.
And what they've learned that byintegrating AI personal tutoring
to every single student in theway they like to learn where
they can ask questions and getassistance in their own pace has
made a dramatic improvement inthe results of all students
taking it.
And it basically took theequivalent of two years.
(20:42):
of learning and was able toachieve the same results in six
weeks.
The students who participated inthis test outperform 80 percent
of other educational solutionsin developing nations, and it
showed benefits topics acrossEnglish, AI knowledge, digital
skills, and so on.
It even had an improvementperformance in other curriculum
(21:04):
subjects outside of the program,just because it gets the
students to be excited aboutlearning and being able to be
more successful.
I said that many times before, Ithink the education system in
the 20th century has made verylittle progress in the last
hundred years.
Most of the things that aretaught are still taught by a
teacher with a board behind himin front of a class with a lot
(21:25):
of people that does not make anysense in the year of 2025.
And while AI represents a verybig risk to that, it represents
the biggest opportunity we hadto really unlock fast learning
that is tailored to the needs ofthe individual, both in the pace
that they need to learn thethings they get stuck on and the
way they like to learn, whetherit's playing games, listening to
(21:45):
stories, reading books, watchingvideos, whatever works for them,
or any combination that will betailored to the specific
individual needs and allowingthem to achieve significantly
better results faster.
And now there's a first researchactually showing it again, two
years of learning progresswithin six weeks of using an AI
tutor, absolutely mind blowing.
And really exciting from myperspective.
Now I told you, we also got totalk about blue collar jobs in
(22:07):
this episode.
Well, a few interesting thingshave happened this week as far
as releases.
So a Chinese company hasreleased a new version of the
robot called black Panther.
So it's black Panther versiontwo, and it can sprint a hundred
meters in under 10 seconds.
So.
As the name suggests, it lookslike a panther, like it looks
like a four legged animal, butit can run crazy fast, meaning
(22:30):
it can run faster than mostpeople on the planet.
This is an Olympic record kindof speed.
The abilities of these robotsare accelerating very, very
fast.
Take a robot like this tohelping in any kind of natural
disasters where they can run onuneven surfaces and get to
places very, very fast.
It's actually relatively small,so it's about two feet high, so
it can crawl under and getaround things relatively easy
(22:53):
and think like, again, a bigpat.
And so.
That's very promising.
There's always the flip side.
This can be used for militarypurposes, and then it can be in
the battlefield and do a lot ofdamage if it's not controlled
properly.
So there's always the good andthe bad with each and every one
of these deliveries.
Unitry G1 is a robot that wetalked about many times before.
Unitree is one of the Chinesecompanies that are developing
the most advanced robots thatthere are out there today.
(23:15):
And they have a big robot calledH1, which was the first thing
that they developed, but theirfocus recently have been on G1,
which is its little brother.
It's about four feet tall andthe price point for it right
now, they're already pre sellingit.
So if you want one, you can goand order one, is 16, 000 for
the base model.
That's Very reasonable for thestuff that it can do.
And what they just demoed isthat they demoed the latest
(23:36):
version of G1 actually runningand jogging across a huge
variety of terrains.
So first of all, it can moveagain, a lot faster than a
previous robot, but it can do iton site slope on very steep up
and down slopes on uneventerrains with rocks and debris
and so on.
And so these robots are becominghighly, highly, highly, highly,
capable and they are becomingreally cheap.
(23:57):
Now, not yet, but we talkedabout this last week, that Elon
Musk is claiming that Tesla isgoing to manufacture hundreds of
thousands of these robots in2026.
So I hear from now inpotentially millions in 2027.
And so the mass production ofthese is just around the corner.
Relating to that is an economistnamed Rubini, who got known as
(24:18):
Dr.
Doom for his gloomy economicforecast is warning about the
rapid development of theserobots their potential impact on
the blue collar job market.
He is.
Estimating that the market ofthese robots will reach 7
trillion by 2050 and that theserobots will start disrupting the
labor market in the next year ortwo.
(24:40):
So again, aligned with thetimelines that Elon Musk was
sharing in his predictions onhow many of these are going to
manufacture.
So if you put in all of thistogether, the capabilities of AI
agents combined with the ofrobots.
And you see a huge disruption toalmost every job out there.
And the pace in which it'spotentially happening.
And yes, we talked about thingsthat will slow it down, but the
(25:02):
pace in which it is potentiallyhappening is a disruption nobody
is ready for.
Now let's dive into some rapidfire announcement on some new
features, capabilities, andother announcement that happened
this week.
First of all, Anthropicannounced voice chat memory
features for Claude.
So the features that are comingand haven't been released yet is
two way voice conversation,similar to what gemini and
(25:23):
Chachapiti already have memoryof past interactions.
Again, something that alsoChachapiti and Gemini already
has and personalized chatcapabilities, which also open AI
already has.
So all these three capabilitiesare coming to Claude.
I must admit I use Claude a lotand I like it for many different
reasons.
One of the things that drives mecrazy is that it still doesn't
(25:43):
have internet access.
And yes, I know you can get itto connect to the internet
through third party tools.
But I don't understand why.
Anthropic hasn't released thatcapability yet.
So if somebody from Anthropic islistening, please add that
functionality, to your toolbecause I love using it, but
it's just really far behind theother tools from that particular
perspective.
Now, another really interestingthing that Anthropic just
(26:04):
launched and that is availableis feature called citations and
citations allowed to ground AIresponses in source documents,
dramatically reducinghallucinations, and it also
provides references to specificsentences and passages in the
original data.
Now it's going to be availablethrough the API to Cloud 3.
5 SONNET and Cloud 3.
(26:24):
5 HYCU and it's going to beavailable through their API or
through Google Vertex where youcan also get access to the
Anthropic platform.
Just as a quick note, you can dostuff like that today In any
chat tool that you're using, Ido this every time I reference
information, every time I ask itquestions about information that
I provided, I ask it for exactquotes, and I ask it for the
(26:45):
name of the document, the page,And the specific paragraph where
they found the quote so I can goand verify the information that
it's giving me.
And that dramatically reduceshallucinations and also allows
me to verify the informationvery, very quickly.
But having it available nativelyin the API would be absolutely
awesome.
Now, similar to this, Perplexityreleased what they call Solar
Pro API, which does similarthings.
(27:06):
It allows you to get citationsand customize sources in the API
to answers that it's giving you.
While focusing on speed andaffordability, that's
perplexity.
So again, the concept ofhallucinations is something
that's going to be reduceddramatically this year as well,
which will give us much bettertrust on these tools right now,
because right now it's a hit ormiss.
(27:27):
And yes, the hit or miss rightnow is probably 80 percent
correct and 20 percentincorrect, but there are many
use cases where that is just notacceptable.
Now staying on the topic of newmodels and capabilities, we
spoke A lot last week about deepseek version three, which came
almost out of nowhere and tooknumber seven on the leaderboard
On the LM sys chatbot arena.
Well, they just released athinking model that integrates
(27:50):
actually the ability to do testtime compute.
So think as it's doing the taskin combination with deep
research capabilities and thatmodel jumped into number four of
the chapel arena, actuallytaking number one on several
different aspects.
I must remind you that thismodel was developed on tens of
cents on the dollar comparedwith the amount of money that
(28:10):
was invested in the leadingmodels from open AI, Anthropic,
Gemini, and so on.
And because of that, they canprovide this kind of capability
for extremely low cost.
Now you can use it through theAPI, you can use it on their
chat, and you can also take anduse it however you want because
it's a full open source modelthat you can actually go and
(28:30):
grab from their website or fromHugging Face, install it on your
servers or on your computer andrun it locally with your data
without sending it anywhereelse.
This is to me is a huge promisefor a future in which we may not
need 500 billion of investmentin data centers, and then the
cooling and power and theenvironmental damage that comes
(28:51):
with it.
Because if they can achievethese kinds of results with
significantly less compute andsignificantly less time, there
is hope that other companies andhopefully will do the same.
And if competition tells usanything, that's what's going to
happen.
And again, that's why I findthis to be very, very good news,
despite the fact that this comesfrom China and not from the U S.
Going back to another piece ofaddition about agentic news,
(29:14):
Microsoft just released AutoGen0.
4 and that provides thecapability to users of the
Microsoft ecosystem to developand integrate agents directly
ecosystem offering custom toolsto build and deploy agents
across enterprises.
Some interesting things aboutthis model is that they're
achieving cost effectiveefficiencies that were not
(29:36):
possible before, while enablingasynchronous processing, meaning
multiple agents can run inparallel, one can do data
collection, one can do parsing,one can do reporting, one can
do, communication and so on, allwhile being orchestrated by a
centralized agent that willbasically be the project
manager.
Another interesting tool thatwas, announced this week is
(29:57):
Mistral, which is a French opensource language model company,
has came out with what theycalled CodeStral 2501, which is
now the most accurate codegenerator out there.
They're achieving 95.
3 FIM accuracy score inprogramming language, surpassing
OpenAI's API by over a millionby 2.
(30:17):
6%, which was the leader so far.
So GPT 01, and it has a contextwindow of 256, 000 tokens, which
allows you to write or review alot of code without model.
I'm reminding you that Mistralis an open source company.
So again, you can go and eitheruse it on their platform with
their API, or just get the opensource and run it locally for
yourself and now develop codebetter than anything else.
(30:41):
It's very, very obvious fromthese tools, as well as
announcements from people likeZuckerberg, that the world of
writing code is going to changeis not going to change, but it's
already changing dramatically.
If you remember, I told you lastweek that Zuckerberg said that
in 2025 coding agents will beable to do the work of mid level
software engineers, and thatthey're already doing it at
(31:03):
Meta.
And two really big pieces ofnews that I kept for the end
because we had too many bigitems in the beginning is iT
spending is projected to hit 5.
5 trillion dollars in 2025.
That's almost a 10 percentincrease for 2024, which was a
record breaking year.
And it's driven mostly by AIhardware needs.
(31:26):
Now Gartner just did a researchand they released it this week.
And they're saying That thegenerative AI is entering, and
I'm quoting the trough ofdelusion meant, which basically
means that there's going to bemore and more expenses and less
and less ROI, Which is astandard phase in the cycle of
hypes and the main concern isthat AI sectors that are showing
(31:48):
insane growth.
but are not really showing bigdifferentiation between every
one of them is raising questionsabout the sustainable value
creation that they can actuallykeep on delivering.
Again, time will tell what isgoing to win, whether it's going
to be this amazing technology orsociety slowing it down.
But it's very obvious to me thatthis is moving forward.
(32:09):
Despite these concerns, doesn'tmean anything about valuations
because literally every singleround that happened to companies
recently has at least doubledthe valuation of companies.
And the previous valuation wasSix to 12 months ago.
So companies valuations aredoubling every six to 12 months,
and that's obviously notsustainable.
Whether the hype is justified,we'll have to wait and see from
(32:31):
everything we're seeing rightnow.
I would say yes, it's too earlyto tell what's gonna be the
actual ROI on that.
And the other piece that isreally, really interesting is
that Google published a researchpaper about what they call the
Titan architecture and It ispotentially the first
significant major advancementssince the transformer paper back
(32:52):
in 2017.
So those of you who don't knowthe story, the paper called
attention is all you need.
That was the paper that Googleresearch have released in 2017
is basically the baseline foreverything AI that we know
today.
And they now released a newpaper called the Titan, which
builds on top of transformer.
So it's not totally replacingthem, but the idea here is
building a neural long termmemory alongside the short term
(33:16):
memory and basing the long termmemory on what they're calling
surprise based learning, whichthey claim is how human
cognitive patterns work.
So basically it's going to lookfor stuff that is out of the
ordinary, that is not alignedwith its existing patterns in
order to consider it new.
And then learn that as a newthing, early benchmarks already
showing superior results toexisting Language models across
(33:39):
several different use cases.
So it'll be very interesting tosee if this really drives the
next cycle of innovation in theactual architecture of how all
these AI models work.
We are going to be back onTuesday with another fascinating
how to episode with a expertsharing with you a specific AI
use case that you can startimplementing in your business
(34:00):
immediately.
If you have not yet ranked andrated this podcast, I would
really appreciate it.
It really helps me get to morepeople and it allows you to play
an important role in deliveringAI education for everyone.
And so please open your appright now, share this podcast
with as many people as you thinkcan benefit from it, and give us
(34:21):
whatever rating and review youthink we deserve.
Hopefully I've earned a fivestar review from you, but
whatever you think, write itdown there, give me feedback.
I actually read all thesecomments.
I would love to hear what youthink.
You can also connect with me onLinkedIn and let me know what
you think about the podcast.
I love chatting with you aboutthe podcast and on things that
you think can make it better anduntil next time, have an amazing
(34:43):
weekend.
Popular Podcasts
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy And Charlamagne Tha God!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Decisions, Decisions
Welcome to "Decisions, Decisions," the podcast where boundaries are pushed, and conversations get candid! Join your favorite hosts, Mandii B and WeezyWTF, as they dive deep into the world of non-traditional relationships and explore the often-taboo topics surrounding dating, sex, and love. Every Monday, Mandii and Weezy invite you to unlearn the outdated narratives dictated by traditional patriarchal norms. With a blend of humor, vulnerability, and authenticity, they share their personal journeys navigating their 30s, tackling the complexities of modern relationships, and engaging in thought-provoking discussions that challenge societal expectations. From groundbreaking interviews with diverse guests to relatable stories that resonate with your experiences, "Decisions, Decisions" is your go-to source for open dialogue about what it truly means to love and connect in today's world. Get ready to reshape your understanding of relationships and embrace the freedom of authentic connections—tune in and join the conversation!