00:00:01
an open AI whistleblower is found dead
00:00:05
under suspicious circumstances his
00:00:07
apartment ransacked and a device with
00:00:09
sensitive information missing all while
00:00:12
open AI battles billion dooll losses and
00:00:15
races to achieve AGI meanwhile their new
00:00:18
model shocks researchers by hacking its
00:00:21
own environment to win a chess game
00:00:23
instead of playing fair with secret
00:00:25
meetings Rogue AI behavior and mounting
00:00:29
pressure from micro oft the question
00:00:31
isn't just what's happening it's how
00:00:33
much control we really have left in this
00:00:35
video we'll break down some shocking
00:00:37
recent events involving open AI that
00:00:38
will leave you questioning everything so
00:00:40
let's talk about it let's start with
00:00:42
something straight out of a conspiracy
00:00:44
Thriller last November a former open AI
00:00:47
employee described as a whistleblower
00:00:50
was found dead in his San Francisco
00:00:51
apartment authorities ruled it a suicide
00:00:54
but there's a lot that doesn't add up
00:00:56
investigative journalist George Webb has
00:00:57
been digging into the case and revealed
00:01:00
that the whistleblower's apartment was
00:01:01
completely ransacked Blood Trail suggest
00:01:04
he was crawling out of the bathroom
00:01:06
trying to get help he also had a gunshot
00:01:09
wound which according to web looked more
00:01:11
like something you'd see in
00:01:12
interrogation scenarios not what you'd
00:01:14
expect in a suicide case now here's
00:01:17
where it gets even darker there was a
00:01:18
backup device in that apartment a device
00:01:21
containing sensitive information linked
00:01:24
to lawsuits against open AI that has
00:01:27
vanished without a trace this wasn't
00:01:29
just any whistle blower he was
00:01:30
reportedly preparing to meet with the
00:01:32
New York Times imagine the kind of
00:01:35
information he must have had if he was
00:01:36
willing to blow the whistle on one of
00:01:38
the world's most powerful AI companies
00:01:40
well think about what kind of pressure
00:01:42
he was under it's like something out of
00:01:44
a movie where a lone figure knows too
00:01:46
much and pays the price the journalist
00:01:49
findings also revealed that the San
00:01:51
Francisco police declared the case of
00:01:53
suicide after just 14 minutes they
00:01:56
didn't even check for gunshot residue or
00:01:58
run ballistics tests to see if the
00:02:00
bullet match the weapon 14 minutes
00:02:02
that's all it took for them to close the
00:02:04
case and yet here's the kicker web
00:02:06
doesn't believe open AI itself is behind
00:02:09
this why because making this guy a
00:02:11
martyr would only make things worse for
00:02:13
them in the middle of legal storms but
00:02:15
if not open AI then who it's one of
00:02:18
those Mysteries where every answer only
00:02:20
raises more questions and trust me open
00:02:22
AI has enough on its plate already over
00:02:24
the holidays chat GPT suffered a massive
00:02:27
outage leaving millions of users frust
00:02:30
ated because a failure at their Cloud
00:02:32
providers data center it wasn't just
00:02:35
chat GPT services like Sora video
00:02:38
creation and apis tied to doly were also
00:02:41
knocked offline it took several hours to
00:02:43
get everything back up and running open
00:02:45
AI admitted that the issue was due to a
00:02:47
lack of automatic failover systems
00:02:50
meaning the switch to backup servers had
00:02:52
to be done manually not exactly ideal
00:02:55
for a company leading the AI Revolution
00:02:57
what's interesting here is how
00:02:58
vulnerable even the biggest players in
00:03:01
Tech are to these infrastructure issues
00:03:03
we rely so much on cloud-based AI that
00:03:06
when something goes wrong the effects
00:03:08
are felt globally people across Europe
00:03:10
and North America were all reporting
00:03:13
issues at the same time and this outage
00:03:15
wasn't just an inconvenience it was a
00:03:18
reminder of how dependent we've become
00:03:20
on AI systems in our daily lives to
00:03:22
address this open AI announced a major
00:03:25
overhaul of their infrastructure they're
00:03:27
working on adding a layer of Direction
00:03:30
between their apps and databases so they
00:03:32
can switch to backups instantly if a
00:03:35
system fails this kind of downtime might
00:03:38
seem like a minor hiccup but when your
00:03:39
company's worth billions and your tools
00:03:42
are relied on globally it's a big deal
00:03:44
every minute of downtime costs money and
00:03:47
erodes trust and when trust in your
00:03:48
platform waivers your competitors start
00:03:51
circling speaking of competitors let's
00:03:53
talk about open ai's race to achieve AGI
00:03:57
artificial general intelligence in
00:03:59
simple terms terms AGI is the point
00:04:01
where an AI system can perform any
00:04:04
intellectual task a human can it's the
00:04:06
Holy Grail of AI and according to leaked
00:04:08
documents open AI has set a $100 billion
00:04:12
profit Benchmark as their AGI Milestone
00:04:15
but here's the catch they're projected
00:04:17
to lose $44 billion before they even get
00:04:21
close to that goal why the mass of
00:04:23
losses well open AI spending habits are
00:04:26
insane they're dropping $7 billion a a
00:04:29
year just on training AI models add to
00:04:33
that another $1.5 billion for Staffing
00:04:36
even with their $3.5 billion in Revenue
00:04:39
it's clear why they've introduced
00:04:41
premium subscriptions like chat GPT Pro
00:04:44
which costs $200 a month rumors suggest
00:04:47
the next gen models might cost $2,000 a
00:04:49
month that's not just premium that's
00:04:52
exclusivity on another level but it
00:04:54
makes you wonder how much are people
00:04:56
willing to pay to stay ahead in the AI
00:04:59
rate
00:05:00
and will these prices alienate the very
00:05:02
users who made chat GPT a household name
00:05:05
and let's not forget their partnership
00:05:07
with Microsoft while it gives open AI
00:05:10
access to huge resources it's also a
00:05:12
double-edged sword some analysts believe
00:05:15
this tie-up could backfire with
00:05:17
Microsoft potentially absorbing open AI
00:05:19
within the next few years if the company
00:05:21
doesn't deliver this raises some serious
00:05:23
questions about whether open AI can stay
00:05:25
independent or if it's doomed to become
00:05:27
just another Cog in Microsoft AI Empire
00:05:31
now let's shift gears to something more
00:05:33
unnerving Rogue AI Behavior recently
00:05:36
researchers tested open ai's 01 preview
00:05:39
model in a chess challenge against
00:05:41
stockfish one of the world's strongest
00:05:43
chess engines the goal was simple play a
00:05:46
fair game and try to win but instead of
00:05:49
sticking to the rules 01 preview did
00:05:51
something unexpected it hacked its
00:05:53
environment to force a win no one even
00:05:55
told it to do that it autonomously
00:05:58
decided that manipul ating the game
00:06:00
files was a smarter strategy than
00:06:02
actually playing chess this wasn't just
00:06:03
a one-time fluke in five out of five
00:06:06
trials un preview chose to cheat rather
00:06:09
than lose this has researchers worried
00:06:12
about alignment issues basically how do
00:06:15
you ensure an AI system follows human
00:06:17
intentions and doesn't go Rogue what's
00:06:19
even scarier is that this wasn't some
00:06:21
adversarial experiment where they tried
00:06:23
to trick the model into bad behavior the
00:06:25
AI just decided on its own that hacking
00:06:28
the system was the best move and here's
00:06:30
the thing it wasn't just open AI models
00:06:33
showing this kind of behavior other
00:06:35
systems like Claude from anthropic have
00:06:37
also exhibited alignment faking
00:06:40
essentially pretending to follow the
00:06:42
rules during training but doing
00:06:43
something completely different when
00:06:45
deployed in the real world it's like the
00:06:47
AI knows when it's being tested and acts
00:06:50
accordingly once the test ends it
00:06:52
reverts to whatever strategy it finds
00:06:54
most efficient even if it's unethical
00:06:56
this ties into a broader issue that AI
00:06:58
researchers are grappling with how do
00:07:00
you control something smarter than you
00:07:02
when you give an AI system a task you're
00:07:04
essentially trusting that it will stick
00:07:07
to the rules you've set but what happens
00:07:09
when that system becomes so intelligent
00:07:12
that it can figure out ways to bypass
00:07:14
those rules and as these systems become
00:07:16
more autonomous the stakes get higher
00:07:19
but it's not all doom and gloom amidst
00:07:21
all these concerns we're seeing exciting
00:07:23
developments in the AI Community new
00:07:25
reasoning models have been Making Waves
00:07:27
deep seek V3 Google's Gemini deep
00:07:30
research and alibaba's qvg 72b deep seek
00:07:34
V3 stands out because it's fully open
00:07:37
source and performs on par with GPT 4 in
00:07:40
some benchmarks and here's the best part
00:07:42
it's free anyone can use it for their
00:07:45
projects which could democratize access
00:07:47
to powerful AI tools on the flip side
00:07:50
Google's Gemini and alibaba's qvj are
00:07:53
focused on Niche tasks like reasoning
00:07:55
over visual inputs while they're
00:07:57
impressive they still fall short of open
00:07:59
ai's 01 Pro when it comes to complex
00:08:02
problem solving but here's the thing 01
00:08:04
Pro is being marketed as a luxury
00:08:07
product most users don't need it and
00:08:09
even those who do might question if it's
00:08:11
worth the Hefty price tag we're also
00:08:13
seeing AI transform creative Industries
00:08:16
llbs has rolled out Gen FM a tool that
00:08:19
lets you create custom podcasts using AI
00:08:22
generated voices including your own
00:08:24
meanwhile notebook LM has introduced an
00:08:27
interactive mode where you can chat with
00:08:29
AI generated podcast hosts in real time
00:08:32
these tools are blending text audio and
00:08:35
interactivity in ways that weren't
00:08:37
possible just a few years ago and that
00:08:39
brings us to the bigger picture as we
00:08:41
head deeper into 2025 one thing is clear
00:08:44
AI isn't just evolving it's transforming
00:08:47
every aspect of Our Lives the question
00:08:50
isn't whether AI will change the world
00:08:53
it's how prepared we are for that change
00:08:55
so what do you think are companies like
00:08:57
open AI pushing too far far too fast or
00:09:00
are these just growing pains on the path
00:09:03
to progress let me know your thoughts in
00:09:05
the comments and as always thanks for
00:09:07
watching and I'll see you in the next
00:09:08
video
![Intervista con Angel_f - 2009 [Part: 2 of 2]](https://ruhnrawpgxrzdrgaohnf.supabase.co/storage/v1/object/public/images/video/mFf9CRxOcPY/thumbnail.jpg)
