00:00:00
so coming in with the story number one
00:00:02
we have a new Chinese humanoid robot
00:00:05
they've honestly been developing these
00:00:06
at quite a rapid pace and now we're
00:00:09
starting to see the different ways in
00:00:11
which they are managing to make these
00:00:12
robots even more flexible we all know
00:00:15
that there is actually an Initiative for
00:00:18
2025 for China to be the leader in terms
00:00:21
of Robotics and today we got limb X
00:00:24
Dynamics which is a shenzen based
00:00:25
startup founded in 2022 and it's
00:00:28
actually you know got success with its
00:00:30
new homid that they are essentially
00:00:32
aiming to rival Boston Dynamic so if you
00:00:35
aren't familiar with this current pose
00:00:37
that it is doing this was something that
00:00:39
Boston Dynamics robot the atlas robot
00:00:41
the very famous one actually did in a
00:00:44
demo video and that was something that
00:00:46
this is trying to replicate so it's
00:00:48
pretty interesting because they can
00:00:50
quickly quickly replicate many of the
00:00:52
things that are going on in the US and
00:00:54
considering the rapid pce technology I
00:00:57
wouldn't be surprised if many of the
00:00:58
robots that we do see as I was saying
00:01:00
it's going to be really interesting to
00:01:02
see how the rapid advancements in
00:01:04
humanoid robot technology manages to
00:01:06
spread overseas we all know that
00:01:08
humanoid robots are going to get a huge
00:01:10
boost in 2025 due to the physics
00:01:13
simulations actually being incredibly
00:01:15
more efficient in terms of their speed
00:01:17
and we're going to likely see humanoid
00:01:18
robots perform in ways that we've never
00:01:20
seen before it's going to be really
00:01:22
interesting to see China also manage to
00:01:24
mass-produce these robots and actually
00:01:26
start to have them working in factories
00:01:28
and considering the fact that in China
00:01:29
CH they manag to have really really
00:01:32
efficient manufacturing it's going to be
00:01:34
quite surprising if they do manage to
00:01:36
pull off some really really effective
00:01:38
robots at an industrial scale that is
00:01:41
going to be something I'm looking out
00:01:42
for because we've already seen some
00:01:44
instances of that so that's going to be
00:01:46
interesting so Point number two is
00:01:48
rather interesting because in the past
00:01:49
couple of days in the AI Community what
00:01:51
I've noticed is that there is a very
00:01:54
very clear thing going on where
00:01:57
individuals are starting to talk about
00:01:58
artificial super intelligence more and
00:02:01
more in a realistic term it's no longer
00:02:03
seen as something that is a fad or
00:02:06
something that is this exaggerated term
00:02:09
that is somewhat clickbait but actually
00:02:11
as a real possibility for future AI
00:02:13
systems and Logan Kilpatrick the Google
00:02:16
Lead product lead for Google AI actually
00:02:19
said that straight to shot to ASI is
00:02:22
looking more and more probable by the
00:02:23
month this is what IIA saw now he does
00:02:26
further expand on this but this is a
00:02:28
significant tweet because because this
00:02:30
is something that we hadn't really seen
00:02:32
before Across The Wider Community the
00:02:34
only person that was really talking
00:02:36
about artificial super intelligence at
00:02:38
the time was of course Ilia SATA and
00:02:41
even remarks or claims of AGI were being
00:02:43
currently disregarded as people who were
00:02:45
far too hyped up on AI hype and AI news
00:02:49
and just seriously in the AI bubble but
00:02:51
right now you can see that other labs
00:02:53
and other individuals are talking about
00:02:54
super intelligence of course this is
00:02:56
referring to Elk's Mission trying to
00:02:58
refer to artificial icial super
00:03:00
intelligence and if this is true it kind
00:03:02
of does make sense considering the fact
00:03:04
that that is quite likely what IIA sover
00:03:06
is pursuing and considering the fact
00:03:07
that Google released their recent model
00:03:10
which is actually also something like
00:03:11
test time compute now them starting to
00:03:13
say that look ASI is on the table it
00:03:15
kind of seems that we're all headed in
00:03:16
one general direction now with
00:03:18
artificial super intelligence he
00:03:19
actually goes on to explain a bit more
00:03:21
he talks about you know how he founded
00:03:23
the company to do a straight shot to
00:03:24
artificial super intelligence no
00:03:26
products and then of course he before
00:03:28
saw this as unlikely to work since if
00:03:30
you get the firew spinning on models you
00:03:32
can actually build a real moot but the
00:03:34
success of scaling test time compute
00:03:36
which of course IL satova saw early
00:03:38
signs of is a good indication that this
00:03:40
direct path of just continuing to scale
00:03:42
up might actually work he says we're
00:03:44
still going to get AI but but unlike the
00:03:46
consensus from four years ago that it
00:03:48
would be the inflection point moment in
00:03:49
history it's likely going to look like
00:03:51
just another product release with many
00:03:53
other iterations and smaller other
00:03:55
options within a short period of time
00:03:57
now if you're wondering what's actually
00:03:58
going to happen in 201 5 I did do a
00:04:00
video covering saman's predictions but
00:04:03
if you do want to know a little bit more
00:04:04
we can take a look at some of the common
00:04:06
themes of course he says artificial
00:04:08
general intelligence this is something
00:04:09
that Sam Alman alluded to in a podcast
00:04:11
well he didn't really allude to he
00:04:13
actually stated it that we're going to
00:04:14
get that in 2025 that of course will be
00:04:17
fascinating he also said AI agents a
00:04:20
much better 40 upgrade which is
00:04:22
definitely long awaited for the people
00:04:23
who sincerely wanted GPT 5 much better
00:04:27
memory this is of course something that
00:04:28
I spoke about before much better memory
00:04:30
is certainly coming a longer context
00:04:32
window is definitely something that he
00:04:34
wants and a lot of people do need a
00:04:36
longer context window allows you to
00:04:37
essentially input more in which is of
00:04:39
course really effective for doing a
00:04:41
variety of different tasks I have to be
00:04:43
honest with you guys using Google's 2
00:04:45
million context length window is
00:04:47
something that is remarkable for a
00:04:49
variety of different tasks a deep
00:04:51
research feature and this is really
00:04:53
interesting because this actually takes
00:04:55
aim at Google's I wouldn't say it's one
00:04:57
of their core offerings but genu ly it's
00:05:00
the most underrated AI tool right now I
00:05:02
made a video on it I made some super
00:05:04
prompts from it and I think it's
00:05:06
probably going to be one of the biggest
00:05:07
tools of 20125 because it's basically AI
00:05:09
agents doing research for you but people
00:05:12
don't know that yet it's just called
00:05:14
Deep research so I think something like
00:05:16
that is going to be really really crazy
00:05:17
especially when open AI release it of
00:05:19
course a better Sora that is of course
00:05:22
so with a better Sora that's going to be
00:05:23
really interesting because one thing
00:05:25
that people don't understand is that
00:05:26
when Sora was the leading model open a
00:05:29
didn't really have any pressure to
00:05:30
release anything better but now that
00:05:32
they're not the leading model we're
00:05:33
actually going to have to see if Sora is
00:05:35
actually better than V2 which is going
00:05:37
to be a really really hard feat so when
00:05:40
Sora gets released it's quite likely
00:05:42
that it will be better than Google's VO2
00:05:44
which when you think about what that
00:05:45
means for the video creation industry
00:05:47
that is going to be something that's
00:05:48
absolutely incredible so that is going
00:05:50
to be something super interesting that I
00:05:52
can't wait to see it also says
00:05:53
interestingly many great updates we have
00:05:55
coming were not mentioned at all or very
00:05:58
little and this is really f fascinating
00:06:00
because samman actually says that many
00:06:01
great updates that they have coming
00:06:03
weren't mentioned at all or very little
00:06:06
so it's quite clear that opening eyes
00:06:07
road map in 2025 is going to be one that
00:06:09
is one is full of surprises so that
00:06:11
means that you know just keep your eye
00:06:13
out for anything that they might release
00:06:15
because it's quite likely that we're
00:06:16
going to get caught off guard again with
00:06:17
a tool or product or a service that we
00:06:20
probably do need but haven't yet
00:06:21
requested and this is something that I
00:06:23
think one of the key things that makes
00:06:24
open AI so successful is that they
00:06:26
actually don't just make aii stuff they
00:06:29
actually make them into really good
00:06:30
products which is why of course when you
00:06:32
use chat GPT they use your interface for
00:06:35
example even when you're using Sora it's
00:06:36
actually remarkably more effective than
00:06:38
any other AI tool out there and that's
00:06:41
the reason that I think this company
00:06:42
will still be successful now
00:06:44
unfortunately for open AI if we're
00:06:46
actually talking about the company being
00:06:48
successful we can actually take a look
00:06:49
at what the Godfather of AI has to say
00:06:52
about open AI it's the Godfather of AI
00:06:55
Jeffrey Hinton actually throws support
00:06:57
behind Elon musk's lawsuit against open
00:06:59
Ai and they are actually supporting Elon
00:07:01
musk's lawsuit seeking a preliminary
00:07:04
injunction that would stop open AI from
00:07:06
becoming a fully for-profit company if
00:07:08
you aren't caught up with the news
00:07:11
unfortunately for open AI they are
00:07:13
trying to become a for-profit company in
00:07:16
order to further the mission and of
00:07:18
course Elon Musk has F an injunction to
00:07:20
stop that because essentially open ey
00:07:22
were a charity and he poured millions of
00:07:24
dollars into said charity and now that
00:07:26
the said charity has taken those
00:07:28
donations and it's turned that into a
00:07:30
for-profit company it's essentially
00:07:32
something that they just aren't happy
00:07:34
about which is of course understandable
00:07:36
now it might just be the fact that Elon
00:07:38
Musk is a little bit jealous that of
00:07:40
course open a are a bit ahead and he's
00:07:41
trying to slow them down but it's very
00:07:43
interesting to see that the Godfather of
00:07:45
AI is now supporting this we can see
00:07:47
that he says openi was founded as
00:07:49
explicitly a safety focused nonprofit
00:07:52
and made a variety of safety related
00:07:54
promises in its Charter and it said it
00:07:56
received numerous tax and other benefits
00:07:58
from its nonprofit status allowing it to
00:08:00
tear all of that up when it becomes
00:08:02
inconvenient sends a very bad message to
00:08:05
other actors in the ecosystem and this
00:08:07
is very true if we have a company that
00:08:09
starts out as a charity and then over
00:08:12
time manages to just say you know what
00:08:14
now let's become a for-profit that is a
00:08:16
very bad precedent for companies because
00:08:18
other individuals will try and use this
00:08:21
structure to essentially ensure that
00:08:24
their company is going to be successful
00:08:25
as well so this is going to have all
00:08:27
sorts of implications if it does go
00:08:29
through through that open is allowed to
00:08:31
do a for-profit conversion that's going
00:08:32
to have to be really technical because
00:08:35
they don't want to open the floodgates
00:08:36
for you know various individuals
00:08:38
starting their own nonprofits and then
00:08:40
turning them into for-profits I'm not
00:08:42
someone who's you know completely
00:08:43
educated on the law but I do know that
00:08:46
when you do allow certain things it
00:08:48
essentially sets a precedent meaning
00:08:50
that other people are going to try to do
00:08:51
the same thing as well and of course
00:08:53
landmark cases do get referred to all of
00:08:56
the time so this is going to be
00:08:57
something that is really really
00:08:58
important I truly am not sure of the
00:09:01
decision of this but it will be very
00:09:02
interesting to see I mean both of these
00:09:04
companies are probably going to have the
00:09:06
best lawyers on either side so it's
00:09:07
going to be super interesting to see
00:09:09
where openi manages to fall now here we
00:09:11
have Sebastian Buck who is someone that
00:09:14
actually worked on the buy series of
00:09:17
models these were the smaller series of
00:09:19
models from Microsoft that basically
00:09:21
showed how important training data was
00:09:24
and in this interview clip he actually
00:09:26
talks about the recent study in a video
00:09:29
which I covered where we actually do get
00:09:31
to see the fact that 01 surpasses
00:09:33
doctors well physicians in terms of
00:09:36
diagnosis in certain domains which is
00:09:38
truly fascinating and it's a remarkable
00:09:41
jump in improvement from GPT 4 and I
00:09:45
think it sets a really interesting
00:09:47
statement for the future because if
00:09:49
these diagnosis tools are that much
00:09:51
better at humans the question I have to
00:09:54
ask myself is at what point does it
00:09:56
become Reckless to not use the AI tool
00:09:59
just recently there was a study that
00:10:01
came out on medical diagnosis comparing
00:10:04
human doctors to gp4 human doctors on
00:10:06
that Benchmark were getting 75% okay so
00:10:09
three out of four were correct gp4 gets
00:10:12
nine out of 10 on medical diagnosis like
00:10:14
this is something that we care about I
00:10:16
care about in the real world you and gp4
00:10:18
is already Beyond human now in that same
00:10:20
study they are already foreshadowing
00:10:22
what happens with o1 and saying that
00:10:24
there are now cases where human doctors
00:10:27
are at 50% and o1 is at 90% % I mean
00:10:29
this is just you know completely
00:10:31
incredible and I'm not going to spend
00:10:32
too long on this because we've already
00:10:34
covered this but I mean just take a look
00:10:35
at the paper it's about like super human
00:10:37
performance of a large language model on
00:10:39
the reasoning tasks of a physician they
00:10:41
tested them in four categories I mean
00:10:43
like you know when you look at you know
00:10:45
a one preview and guys this is just 01
00:10:48
preview okay this is not like the the
00:10:49
second iteration of the model just
00:10:51
imagine the third iteration of 01 the
00:10:53
fourth iteration with test time compute
00:10:55
I mean how accurate is that model going
00:10:57
to be especially if it's fine tuned on
00:11:00
you know medical diagnosis and stuff
00:11:01
like that so I mean it's going to be
00:11:03
really interesting especially when you
00:11:05
can have a AI That's able to generate
00:11:07
synthetic data and then train on that
00:11:09
synthetic data so even if there is a
00:11:11
super rare case the AI is able to you
00:11:13
know pull it from its training data that
00:11:15
wouldn't exist in the real world that
00:11:17
really would have doctor stumped so I
00:11:19
think this is genuinely a silent
00:11:20
Revolution that's going on cuz when we
00:11:22
even check right here you can see 01
00:11:23
preview in terms of the score out of 100
00:11:26
I mean Physicians and GPT 4 you can see
00:11:28
the kind of results they get an 01
00:11:30
preview okay preview is literally all
00:11:33
the way up there I mean it's it's truly
00:11:35
remarkable um on on Landmark diagnostic
00:11:37
cases as well we can see 01 preview
00:11:39
performs nearly at 100% it's just
00:11:41
something that is uh yeah pretty
00:11:42
incredible stuff I would say that this
00:11:44
is a good sign though for anyone you
00:11:45
know with kind of health issues because
00:11:46
I think right now we're just getting the
00:11:48
diagnosis but eventually on the tail end
00:11:50
of this we are going to get some
00:11:51
probably significant medical
00:11:52
breakthroughs by the end of this decade
00:11:54
or by the end of next decade certainly
00:11:56
so it's going to be really interesting
00:11:58
at a time for PR erity hopefully now
00:12:00
interestingly there was this model
00:12:02
called Deep seek V3 and this is
00:12:05
something that I do want to say
00:12:06
essentially shook the AI Community this
00:12:09
was because this is a model that
00:12:12
basically managed to perform on par with
00:12:15
a lot of the current llms but that
00:12:18
wasn't the only thing that this model
00:12:19
managed to do this model currently
00:12:21
managed to essentially get the same
00:12:24
benchmarks at a fraction of the cost
00:12:26
I've heard that the cost of this model
00:12:28
is around $20 times cheaper than GPT 4
00:12:31
and from my sources many people are
00:12:33
currently switching for this to their
00:12:35
API Usage Now claw 3.5 Sonet Still
00:12:38
Remains number one in terms of the
00:12:40
coding capabilities but from what I'm
00:12:42
hearing is that many people are now
00:12:43
switching to this model in terms of
00:12:45
their GPT 4 applications now this is
00:12:48
something that is pretty good because if
00:12:50
you've ever wanted to build an AI
00:12:52
applications and maybe the I guess you
00:12:54
could say API costs were something that
00:12:56
were a factor now dsek V3 is coming in
00:12:59
as a very cheap alternative that allows
00:13:02
you to rapidly prototype what you want
00:13:04
without facing heavy inference costs so
00:13:07
this is going to be a true shake up to
00:13:09
the market and there was also some
00:13:11
information regarding this as well that
00:13:14
was a little bit controversial so
00:13:16
essentially if you asked this model what
00:13:19
it is it said I am a large language
00:13:21
model developed by chat gbt that is what
00:13:24
the model would output now of course
00:13:26
that doesn't really make sense if the
00:13:27
model is deep seek V3 it should say I'm
00:13:29
deep seek V3 developed by you know the
00:13:31
company deep seek so um there was some
00:13:33
information going on out there that
00:13:35
basically said that they trained on open
00:13:37
ai's models output which is of course
00:13:40
against open ai's terms of service the
00:13:42
only thing is I'm not sure how open AI
00:13:45
would even recognize that a company is
00:13:47
able to do that I'm sure they've got
00:13:49
Advanced ways to do that but considering
00:13:51
the state of AI right now I'm not sure
00:13:53
that's something that you know open AI
00:13:56
would even want to pursue at the moment
00:13:57
but we do take a look at the benchmarks
00:14:00
you can see here that it does seem to
00:14:02
basically you know outperform GPT 40 on
00:14:05
a variety of different benchmarks and
00:14:06
and you know if we're being honest about
00:14:08
what GPC 40 is currently it isn't the
00:14:11
best model in terms of like you know an
00:14:13
llm that you could use you know Google's
00:14:15
newer model is better I actually use
00:14:17
Google's model every day now um of CLA
00:14:19
for coding you've got claw 3.5 Sonic
00:14:21
that model is you know better in coding
00:14:23
aspect so it's something that just
00:14:26
manages to you know put opening eyes
00:14:28
leadership of course in a state of
00:14:30
contention because we have a model that
00:14:33
is cheaper and basically on the same
00:14:34
level so people are always trying to
00:14:36
find ways to save money and if you can
00:14:38
you know find a way to save money by
00:14:39
using this of course that is something
00:14:41
that people might do now whether or not
00:14:42
people are going to go to the hassle of
00:14:44
actually going to a deep sep3 account
00:14:45
because you know what openi are
00:14:47
essentially doing you know they've got
00:14:48
advanced voic mode they've got a variety
00:14:50
of different other things that still
00:14:51
make them a very good product to use I
00:14:53
still think that the market has allowed
00:14:56
shakeups that are really fascinating now
00:14:58
in addition we ALS also do have this
00:14:59
thing called reproduce which is where
00:15:02
there's stepbystep blueprints that
00:15:04
reveal how to recreate opening eyes
00:15:06
reasoning abilities from scratch so
00:15:08
there's a paper that actually presents a
00:15:10
road map to reproduce the 01 model using
00:15:13
reinforcement learning focusing on
00:15:15
policy initialization reward design
00:15:17
search and component so this is going to
00:15:19
be interesting I'm actually going to be
00:15:20
covering this in a video probably
00:15:21
tomorrow but this is fascinating because
00:15:24
if the secret of this is out it
00:15:26
basically means that the secret of AGI
00:15:28
is out and if that is the case and that
00:15:30
means we are about to get AGI globally
00:15:32
and if we're about to get AGI globally
00:15:35
then that means that there's about to be
00:15:37
the acceleration that we've never seen
00:15:38
before because none of these countries
00:15:40
want to lose the AI race now if we're
00:15:42
talking about you know large language
00:15:43
models and different architectures and
00:15:45
of course the o1 architecture there is
00:15:48
this fascinating paper from meta called
00:15:51
LCM large concept models and with large
00:15:54
concept models this is essentially a
00:15:56
kind of model that basically tries to
00:15:59
make the model think how humans think so
00:16:01
you know how you know large language
00:16:03
models think in you know it's tokenized
00:16:05
and just the tokenize of stuff basically
00:16:07
um you know this is probably a better
00:16:09
explanation um this is an AI researcher
00:16:11
and he said that I'm more and more
00:16:12
confident that tokenization will be gone
00:16:15
because of course humans don't think in
00:16:16
tokens tokens are hard quoted
00:16:19
abstractions in llms that lead to weird
00:16:21
Behavior Uh you know llms can solve PhD
00:16:24
math questions but canot answer is 9.9
00:16:27
bigger than 9.11
00:16:29
uh So Meta is Shifting you know LMS to
00:16:31
lcms which are you know changing from
00:16:34
next token prediction to next concept
00:16:36
prediction and the concept is treated as
00:16:38
a sentence you know representing an idea
00:16:40
representing an abstract idea or action
00:16:42
and I think this is something that's you
00:16:44
know really interesting to see if it
00:16:45
gets embedded into other architectures
00:16:47
because we all know that the
00:16:49
tokenization feature it not feature you
00:16:51
know it's it's a part of the
00:16:52
architecture that is something that has
00:16:54
caused problems before and we've seen
00:16:56
meta previously tackle these kinds of
00:16:58
things where they're looking at things
00:17:00
different to everyone else so I think if
00:17:01
meaan of course continue to push that
00:17:03
it's going to open up a really nice area
00:17:06
in terms of AI research because we're
00:17:07
going to get areas that most people
00:17:09
currently aren't focused on and this is
00:17:10
something that's really important
00:17:11
because if you want to get truly Smart
00:17:13
Systems having them thinking in
00:17:15
different ways or ways that are really
00:17:16
similar to us is going to really advance
00:17:18
that now we also got this clip that has
00:17:21
been doing the rounds on Twitter and
00:17:22
this is where we get the exop AI
00:17:24
employee Daniel kokalo actually
00:17:27
explaining what AGI looks like when it's
00:17:30
going to be released so this is a super
00:17:32
fascinating interview because I don't
00:17:34
think there have been many individuals
00:17:36
from Frontier Labs actually releasing
00:17:39
the kind of information we get in this
00:17:40
short one minute clip and this is
00:17:43
something that is fascinating to me
00:17:44
because one of the things as well is you
00:17:46
know it kind of coincides with what Sam
00:17:48
Alman said previously where he basically
00:17:50
said that you know we're going to get
00:17:52
AGI and it's just going to be like
00:17:54
another day for the average person so
00:17:56
take a look at what he says here what
00:17:58
you would see eventually is uh some sort
00:18:01
of very exciting announcements and some
00:18:04
sort of very exciting products being
00:18:06
rolled out and things like that uh into
00:18:09
the world but by by the time that's
00:18:11
happened uh you but by the time you
00:18:13
actually see the plumber uh the the
00:18:14
robot plumber walking around AGI
00:18:16
research will have been automated for
00:18:18
like a whole year or more and the things
00:18:22
living on the data centers will be super
00:18:24
intelligent like literally across the
00:18:26
board qualitatively quantitatively
00:18:28
better than the best humans at
00:18:29
everything um and they won't just have
00:18:31
been designing robot plumbers like
00:18:33
there's a million of them they'll be
00:18:34
working on lots of things at the same
00:18:35
time they'll be the robot plumbers but
00:18:36
there'll also be new weapons new drones
00:18:39
there'll also be all sorts of new
00:18:41
software there'll also be all sorts of
00:18:42
new basic science that will have been
00:18:44
figured out new types of physics new you
00:18:46
know chemistry new biology um it'll be
00:18:49
this giant explosion of amazing new
00:18:52
technologies that will be sort of like
00:18:54
the the scientific and Industrial
00:18:56
revolutions like if you think about what
00:18:58
happen in 500 years from like 1500 to
00:19:01
2000 it's just this insane explosion of
00:19:05
uh intellectual progress in science and
00:19:08
then also like physical progress in
00:19:09
terms of like machinery and and um Tech
00:19:12
you know tools and stuff like that I'm
00:19:15
imagining a similar sort of explosion
00:19:16
happening over the course of maybe two
00:19:18
years uh all under the control of uh one
00:19:21
AI project probably so for those of you
00:19:23
who are wondering how AI gets here that
00:19:25
is going to be something that is super
00:19:27
intriguing to me because of course I
00:19:29
would love to know what life is like
00:19:31
when AI gets here and of course we do a
00:19:33
lot of speculation on this channel so
00:19:35
it's going to be super interesting to
00:19:36
see if his predictions actually do come
00:19:39
true in the sense that we're going to
00:19:41
start to see a variety of different
00:19:43
inventions and you know products getting
00:19:44
released and then if we're actually able
00:19:47
to tell if of course AI research has
00:19:49
been automated now interestingly enough
00:19:52
gearing up into 2025 we can see that the
00:19:55
Google CEO Cinder Pai has actually told
00:19:57
employees to gear up for a big 2025 as
00:20:01
the stakes are high now this is super
00:20:03
interesting because we've seen that in
00:20:05
the last few weeks in the last quarter
00:20:08
of 2024 Google really showed everyone
00:20:12
that they are still the kings of this AI
00:20:14
stuff and I think this was surprising to
00:20:17
many people in the AI industry because
00:20:19
what we saw from Google before was
00:20:21
rushing to get products out we saw
00:20:23
sloppy releases we saw delayed releases
00:20:26
and we even saw you know releases that
00:20:28
just didn't happen altogether at
00:20:30
Google's IO conference we saw coming
00:20:32
later this year more times than we had
00:20:34
actual product releases but with the
00:20:36
recent Google vo with a recent new model
00:20:38
with a recent thinking model I mean it's
00:20:41
been absolutely an onslaught of
00:20:43
different products that we've had it's
00:20:45
been nothing but gems that we've
00:20:47
literally had with deep research I mean
00:20:50
the list just goes on and on and on
00:20:52
Google have just you know raised the bar
00:20:54
to I think definitely the gold standard
00:20:56
now in terms of across the board AI in
00:20:59
terms of usability so it's going to be
00:21:01
super interesting to see where they
00:21:02
stand now if you actually read further
00:21:05
into this article we can see that you
00:21:06
know it talks about how 2025 is going to
00:21:09
be a critical year and it's important
00:21:11
that they internalize the urgency of
00:21:13
this movement and they need to move
00:21:15
faster as a company as these Stakes are
00:21:17
high and that these are disruptive
00:21:19
moments in 2025 they need to be
00:21:21
relentlessly focused on unlocking the
00:21:23
benefits of this technology and solve
00:21:26
real problems now this is good for us
00:21:28
because of course it means that we're
00:21:29
going to be getting some killer products
00:21:32
and like I said before a lot of the
00:21:33
things that they've released have been
00:21:35
really tremendous in terms of my
00:21:37
productivity and just downright the
00:21:39
things I've been able to do you know
00:21:40
I've spoken about the fact that deep
00:21:42
research is an amazing feature also the
00:21:44
fact that you know the agent browser
00:21:46
when I've seen the demos of that it
00:21:47
looks really really amazing and I can't
00:21:50
imagine the amount of time I'm going to
00:21:51
save with that so it's going to be super
00:21:53
interesting because of course you know
00:21:55
if you thought that 2024 was crazy they
00:21:58
are saying that 2025 is the critical
00:22:01
year so if you're someone who's a Google
00:22:03
Fanboy or maybe even an open AI Fanboy
00:22:06
definitely keep your eyes on Google
00:22:08
because they're going to have a lot of
00:22:09
products that are coming out and so here
00:22:11
you can see that they actually said they
00:22:13
know that they have some work to do in
00:22:14
20125 to close the gap and establish a
00:22:16
leadership position there as well but of
00:22:18
course they know that you know that they
00:22:20
want to be number one you can see right
00:22:22
here that they said that I expect some
00:22:24
back and forth in 2025 but I think we be
00:22:28
stay to of the art so Cinder Pai is
00:22:30
basically stating here that in 2025 he
00:22:33
expects them to actually be the number
00:22:35
one AI company in terms of overall
00:22:37
market share in terms of just how good
00:22:40
everything is and honestly by the end of
00:22:42
2024 we could argue that Google is
00:22:44
number one because when we looked at the
00:22:45
benchmarks when we looked at the product
00:22:47
releases I mean the stuff they did was
00:22:50
really really impressive so you know
00:22:53
right here they're stating that you know
00:22:54
he acknowledged that Google has to had
00:22:56
to play catchup but he says in history
00:22:58
you don't always need to be the first
00:22:59
but you have to execute well and be
00:23:01
really best inclass as a product and in
00:23:03
2025 that's all it's all about so for
00:23:06
you consumers out there you know don't
00:23:07
sleep on your Google devices just yet
00:23:09
I'm sure there's going to be some major
00:23:11
updates coming as well and of course
00:23:13
with this we do have this individual
00:23:16
jyang Ling who is you know building quen
00:23:18
models at Alibaba actually responded to
00:23:21
this tweet where someone said they want
00:23:23
a Sonic level 70b model which is
00:23:25
essentially just a smaller large
00:23:27
language model like chat GPT someone
00:23:28
said that they wanted a Sonic level
00:23:30
version of that and he said wait on me
00:23:32
so I mean it's going to be really
00:23:34
interesting to see if we manage to cram
00:23:36
more intelligence into these smaller
00:23:38
models a 70b clo 3.5 Sonic level model
00:23:42
I'm not sure if they can pull that off
00:23:44
because I think some things are just
00:23:46
achieved through skill and I think even
00:23:48
if you try to be be as efficient as you
00:23:50
can you might not be able to just get
00:23:52
that kind of intelligence out of the
00:23:53
model from that size if you guys know
00:23:55
what I mean but then again we've seen
00:23:57
you know a variety of different things
00:23:59
become more efficient become faster so I
00:24:01
wouldn't be surprised if this is
00:24:03
something that occurs and I'm never
00:24:05
going to bet against the trend of AI
00:24:07
because I've done that in the past and
00:24:08
it doesn't work well and AI is an
00:24:10
exponential Trend so by the end of 2025
00:24:13
we could be in a remarkably different
00:24:14
position now in addition what we also
00:24:17
got this week was something rather
00:24:19
interesting we finally got the
00:24:20
definition of AGI and this one was
00:24:22
really good because we now know that
00:24:24
open AI have defined the AGI as a system
00:24:27
capable of generating $100 billion in
00:24:30
profits now I don't think this is like
00:24:32
you know the specific definition of AGI
00:24:34
I think this is more like a corporate
00:24:36
cookie cutter term just to you know
00:24:38
please the individuals at Microsoft
00:24:41
because opening eyes deal with Microsoft
00:24:42
is you know it's kind of on Shaky Ground
00:24:44
based on all the things going on but you
00:24:46
know they're negotiating the agreement
00:24:48
and I think Microsoft of course wants of
00:24:50
course to retain control over some of
00:24:53
this AI stuff but it's going to be super
00:24:55
interesting to see where things head
00:24:57
with them because of course I think that
00:25:00
like a lot of people are starting to say
00:25:02
that you know you've compared Microsoft
00:25:04
and you've compared Google but Microsoft
00:25:07
doesn't have their own models so like
00:25:09
once Microsoft's contracts end with open
00:25:11
AI what are they really going to do as a
00:25:14
company and that is of course you know
00:25:16
something that brings us to the question
00:25:17
but it's not that crazy it's not that
00:25:19
big of a deal I'm sure they'll work out
00:25:21
some agreement and one of the last
00:25:22
stories we have here is actually super
00:25:24
interesting so so we actually have this
00:25:26
video of a humanoid robot released by by
00:25:28
matrix 1 and this one's rather
00:25:30
fascinating and maybe other companies
00:25:32
can take a leaf out of this book because
00:25:34
I'm more intrigued about this company
00:25:36
than I am any other company so this is a
00:25:38
company that I don't even know the name
00:25:40
of I'm I'm guessing it's called Matrix
00:25:42
and they've released this robot that
00:25:43
I've only seen one video on so the
00:25:45
captions you're seeing on screen I
00:25:47
didn't make those I just found this
00:25:49
video on Twitter I couldn't find the
00:25:50
original video I googled this robot
00:25:53
nothing has come up so far but the
00:25:55
technology in this robot does look
00:25:57
futuristic it does look really Advanced
00:26:00
I mean I haven't really seen any kind of
00:26:02
humanoid that looks like this I think
00:26:04
the last robot I saw that looks like
00:26:05
this mildly is the you know Honda Asimo
00:26:09
and that was over perhaps 20 years ago
00:26:11
but this robot right here it literally
00:26:13
looks like it's from the future and the
00:26:16
fingers look really flexible in fact if
00:26:18
I had to say it it kind of resembles
00:26:20
maybe an astronaut suit suit like the
00:26:22
SpaceX one but there isn't much to go
00:26:24
off here other than this looks really
00:26:26
effective I mean if this is able to do
00:26:28
some remarkable tasks it's going to
00:26:30
definitely feel like the future I do of
00:26:32
course wonder how much this kind of
00:26:34
product is going to cost but of course
00:26:36
you know you can keep up with the
00:26:37
progress with the account on Twitter
00:26:39
that manages to cover all of that stuff
00:26:42
at Cyber Robo as I will leave a link to
00:26:44
that stuff as well and hopefully you
00:26:45
guys have had an amazing week and
00:26:46
hopefully you guys have had an amazing
00:26:48
New Years as we go on into 2025
![Intervista con Angel_f - 2009 [Part: 2 of 2]](https://ruhnrawpgxrzdrgaohnf.supabase.co/storage/v1/object/public/images/video/mFf9CRxOcPY/thumbnail.jpg)
