STANDARD EDITION: Attacking Signal, Blogging Getting MORE Important, AI's Final Form, Claude 3.7 vs. World, Censorship as a Service, and more...
➡ Protect Against Bots, Fraud, and Abuse. Check out WorkOS Radar at: workos.com/radar
You are currently listening to the Standard version of the podcast, consider upgrading and becoming a member for the full version and many other exclusive benefits here: https://newsletter.danielmiessler.com/upgrade
Subscribe to the newsletter at:
https://danielmiessler.com/subscribe
Join the UL community at:
https://danielmiessler.com/upgrade
Follow on X:
https://x.com/danielmiessler
Follow on LinkedIn:
https://www.linkedin.com/in/danielmiessler
Become a Member: https://danielmiessler.com/upgrade
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
S1 (00:00):
Does your app get fake sign ups, throw away emails
or users abusing your free tier? Or worse, bot attacks
and brute force attempts? Work OS radar can block all
of this and more. Simple API gives you advanced device
fingerprinting that can detect bad actors, bots, and suspicious behavior.
Your users trust you. Let's keep it that way. Check
(00:23):
out work OS Radar at work OS comm slash radar.
That's work os.com/radar. Unsupervised learning is a podcast about. Trends
and ideas in cybersecurity, national security, AI, technology and society
and how best to upgrade ourselves to be ready for
what's coming. All right. Welcome to unsupervised Learning. This is Daniel,
(00:51):
episode 470. Let's get into it. Reading a couple of
new books in addition to this month's books for UL.
So the UL books are 1984 by Orwell and Animal
Farm by Orwell. And these two books I just added
to the list. I'm about done with the Technical Republic
(01:11):
and about to jump into money, lies and God. And actually,
there's another one I'm going to add to this list,
somebody who came on the Goodfellas podcast. So I'm going
to go find his book as well. But, um, yeah,
having a good time with some reading. Got a friend
in Guatemala who was recently laid off senior engineer, focused
(01:33):
around monitoring solutions, but really can do lots of different stuff.
So I recommend you check out his LinkedIn and see
if if you might have anything. My friend Monica is
offering 25% off her Security Leadership Masterclass, which I consulted for,
and I think it's really good for people trying to
get into leadership around security. I got a funny meme
(01:56):
here about, um, how to calm down. A buddy who
loses a bunch of money in crypto. It's just. It's
just funny. Nothing useful there except for laughter, which is useful.
LinkedIn posts about my ultimate app, which I keep iterating on.
I'm going to go ahead and click through to this one.
So this is essentially like the main app that I've
(02:19):
been building for the longest time. And I've got like
different versions of it. I've got like a commercial version,
I've got the overall arching one, but essentially it takes
inputs from any medium, from any platform and does a
ton of analysis to determine how good it is. Okay,
so just imagine that this is like the most important
thing to me is most important like current buildable use
(02:44):
for AI. So all this stuff is happening in the world, right.
All this news, new intelligence is is dropping open source intelligence,
recon vulnerabilities are popping up on websites. Um, my favorite
thinkers are releasing a new blog, blog post, a new essay,
a new article, a new video on YouTube comes out right.
(03:06):
All the authors that I follow, they're releasing new books, right? Um,
comments are being made which might reveal something really, really interesting. Uh,
new art is coming out, new books, new movies that
I should be watching. Um, comments from my friends, new
companies being launched. All this stuff is happening, you know,
(03:28):
millions of things per day. Hundreds of thousands, tens of
thousands or whatever, uh, pieces of content coming into my
RSS feeder. And all of that is, is something I
could potentially benefit from. Okay. So this app that I've
been working on for the longest time, really starting, uh,
(03:49):
beginning of 23, is essentially the consumption mechanism for this.
Pull it all together, then use AI to run it
through a filter. Okay. That filter. And I've already built
this way back then. The filter basically does a quality
check on all of these things to see if it's
(04:09):
full of interesting ideas, if it's creative, if it's novel,
if it's surprising, and all these various different aspects similar
to the label and rate, um, fabric pattern that I
have out there that's public. This is just like a
much more advanced version of it, with a lot more
labels and stuff like that, because based on that, the
(04:32):
AI can then or a different AI or a different
part of the same AI can then sort what to
do with it, right? Because it might be an alert.
It might send me an alert if something is important enough, like, hey,
a new vulnerability was found on one of your websites,
or a new vulnerability was found on a bounty program
that I'm monitoring. And by the way, the AI is
(04:56):
telling me my AI is like, oh, by the way,
I found the vulnerability. I found how you could actually
exploit it. I wrote that up into a report and
I submitted it to the thing. So we are waiting
to see if we're going to get paid from this bounty. Right.
So that that's a cool thing. And it might send
me an email which tags it with a certain thing
that says awaiting for bounty payout or something like that.
(05:19):
Or it might be that I just want to know if, um,
this particular company buys any new companies, or I want
to know if this particular company does anything to, like,
raise money. All of these could be like alerts that
I have set up. But it starts with the consumption
of the world and the processing through this AI. Now
(05:42):
I've already built this I've got is probably a version
like three or something, depending on how you call the versions.
But what I want to mention about this is that
the real version of this, the real version of this,
which I've talked about in my other videos, especially the
predictive path AI. Video is that the real version of
(06:04):
this is my AI buddy, my AI assistant, my AI friend, my.
AI my digital assistant. Okay, my universal primary digital assistant
is going to be in charge of doing this for me. Okay.
My APIs that are available to me are the ones
that are doing this processing right. These agents, these automation workflows,
(06:28):
these things that are just running on my behalf. They're
out there in the world collecting, analyzing, assessing, um, labeling,
doing all those things for me. But the quarterback for
all of this is my Da. My Da is monitoring
how many feeds I have, if they're the right feeds,
if this feed like we thought it was awesome, but
(06:49):
it's actually garbage. It might have the permission to go
in and clean that up and fix it. Right. So
the future of this is I don't really ask any
questions about this. I'm not like typing very much. I'm not, um,
really manually interacting with this too much at all. Because
if I want a new source, I say, hey, add
(07:10):
this to my source, okay? While I'm looking at a
page on the screen, hey, add this to my our
list of sources in my says in my ear, yeah,
no problem. Just added it. Okay, I'm watching a new
YouTube video, for example. I'm like, yeah, add this to
our sources and it's like it runs the API command
to go add that. It's now part of the workflow.
(07:33):
It's now in the system right now. How could it
do that? Well, it's going to do that because it's
watching all my screens or it's watching with the camera.
However it's doing it. There's going to be lots of
different ways. But the point is, my Da will be
part of my life. Um, Apple is talking about having
cameras on its AirPods. I don't know if that will
be the next version. Probably not. It'll be the version
(07:55):
like 2 or 3 versions after the current version, but
I'm wearing them now. Like if if these were like
over here and in the back, it had a camera
that was pointing behind me. Amazing. Obviously it should have
a camera that's pointing forward, pointing forward, pointing behind me
with full access to my life context. Full access to
(08:17):
my screen. It should also have access within the screen
as well, so we could look at processes and stuff
like that. But let's just say you could only just
see the screen. I could still ask it things and
it could still do lookups. It could still summarize what
I'm looking at, for example. So this is really the interaction.
The interaction becomes me just talking to my AI and
(08:39):
my AI giving me back results. Now the second level
of this, which comes a little bit later because the
technology just is not quite there yet, but it's more
like the meta glasses. Okay, the meta Metaclasses. As you
saw in the newsletter last week, or maybe it's this
week and we're about to talk about it. The Metaclasses
(09:00):
have sold 2 million units. They are the direction that
things are going. I imagine Apple is going to move
in this direction as well, and kind of offload a
bunch of the processing to the phone, but the next
version of this app that I'm building, the reason I'm
building all these services that are working for me, coming
through my digital assistant, which also doesn't fully exist yet. Okay,
(09:24):
that's also a place everyone is pushing into, including Apple.
But the real thing is that I am watching. I
am talking to somebody and they're like, hey, you know,
you should get in on this deal and blah blah, blah.
And it's crypto related and it's a meme coin. And
I'm just like, uh, well, not interested. Don't want to
(09:45):
hear about that. Um, but let's say I was interested
and let's say it did sound good. In the meantime,
my Da is calling all my APIs. Okay, I have
an API for researching someone. I have an API for
creating a dossier on somebody. I have an API for monitoring,
watching someone's face and determining characteristics of lying. So while
(10:10):
I'm looking with my apple glasses, which is just like
they look like regular glasses, this is pretty far in
the future. Like five years maybe, maybe shorter, maybe like
three years with meta or something. Who knows? But let's
just say three to 5 to 10 years. Okay. As
you're watching, I have this little dial in the corner
of my glasses and it's like, it's like a bullshit meter, right?
(10:33):
It's like, yeah, no, don't believe that. And it could
even be printing out little thing in the text or whatever.
That's like that claim he just made is incorrect. That
was kind of bullshit. Oh, that was kind of smart
or whatever. You could do whatever. You could have an
outline around them based on the type of person you're
looking at. I talked about this in that blog post,
(10:54):
which you should go check out. Actually, here's the video
for it. So you've got to think about what is
possible from your Da when your Da can also control
what you are seeing. The overlay on top of the world. Okay,
so when I walk into a Starbucks and I'm looking
for a new partner, I'm looking for a girlfriend, which
(11:16):
I'm not, by the way, but let's say I were, um,
there could be different outlines based on what, um, the
girls in the Starbucks are broadcasting in their demons. They're
broadcasting that they're creative. They're broadcasting that they like programming,
they're broadcasting that they want to start a business. They're
broadcasting whatever. And my Da could then place different outlines
(11:41):
around them based on highlighting that for me, knowing what
I'm looking for. Right. And more importantly, there's going to
be 1000 or 1 million companies out there pitching to
my Da that their filter on reality is the best one,
because they have the coolest outlines, and they have the
coolest little animations that go around people who are like
(12:02):
artists or engineers or whatever. So this is kind of
these are the steps. This is what I'm building. Like
the stuff I put in that book from 2016, the
stuff I've been talking about in these posts, like I'm
not waiting for it to happen, I'm building the pieces
right now. Like I can't build the hardware. Like I'll
never mess with that. I have to wait for that.
(12:24):
But I can maybe build the assistant, or at least
build something to go on top of the assistant. And
I can absolutely build these services. And this is like
the most important service. This is the gathering and filtering service.
So that is what that's what I've made. Oh and
by the way I the commercial version of this that
(12:44):
I already put out, let me just click this and
see like what we get here. This is my actual threshold.
This is my personal feed for threshold, and I've actually
never showed this. Have I never showed this? I don't
think I've ever showed this before. Let me show you
what this is like. So this is my threshold. And
look what I can do in here. I have a
(13:07):
score here that I can, uh, set the minimum threshold
of quality that has to. The content has to be.
So I have thousands of things like. This is not theory.
I've already built this. Thousands of things are coming into
this thing being parsed, being filtered, being labeled, being rated. Right.
(13:28):
Watch this. It's at 85. I slide this down to 60.
Look I have all these categories as well I hit save.
Look what it does over here. What it just readjusted.
It just readjusted. And watch this watch this view analysis
I have summaries of the contents. You could decide if
you actually want to go and read it, or if
(13:49):
you actually want to go and watch the video and
look at this surface level, middle level, deep level, hidden
message analysis, the ideas that are actually in here, the
recommendations that came out of the content. This thing helps
people save time, going from like hundreds of thousands of
feeds down to a small number of feeds. And even
(14:13):
when you get the small number of feeds that pop
up in this list right here. So let me just
take this to like 95 save. So even when this
thing pops up, I could still decide if I want
to read it or not based on this I usually
keep it around middle. So three ideas, a little review
of it, recommendations coming out of it. It's insane. And
(14:36):
then I always have the link to go click it
and read the actual full thing. So this thing saves
me like infinite time. And I find so much content
that I would never find before. And here, here's the
ultimate idea for this. And this is the reason I'm
building this entire system separate from this app here in
the commercial app. I don't care if the best idea
(14:58):
came from like a 13 year old girl in Nairobi,
because her idea might be as good or better than
Marc Andreessen's latest idea that just came out. But Marc
Andreessen is going to get all the press. It's going
to be through all my feeds. So if I just
rely on that, it's going to be hype cycle based
(15:19):
on social dynamics. So it's going to be surfaced to me,
but I am blocking that out. My algorithm does not
care about how popular that person is. It doesn't care
how about how popular that it turned into being going viral?
It doesn't care. It's judging the quality of the ideas
(15:40):
and the the novelty of the ideas and the creativity.
That's what's determining whether or not I'm going to see
a thing. So that's like the commercial version of this basically,
which is already out. And of course I'm going to
improve that as well. But it's kind of like not
the main point. The main point is this right here,
(16:00):
my interface to the world. This is why I talk
about augmentation okay. Augmentation is this is this is why
I am so excited about, I think of how much
smarter I can be about the world if I'm being
constantly fed the best ideas from the entire planet, the
(16:21):
more stuff I can find from unknown people. And by
the way, when I when I hear from unknown people,
I give them a shout out. I broadcast them out
on on X or on blue Sky or in the newsletter,
or I reach out to them and ask them if
they want mentoring. Like it is so fun to find
(16:45):
like nascent talent in the world and like, try to
give them confidence and lift them up and kind of
broadcast them out. And you've seen me do it a
million times, and it's just it's unbelievable to connect with
people who have interesting ideas. And that's what I love
about this thing is it finds interesting ideas regardless of source.
(17:06):
Forget the source. I don't care about the source. I
want to talk about ideas. All right, so that was that. Um,
the discovery section in the newsletter is getting absolutely insane. Uh,
there's going to be a cutoff. Um, in this episode, um,
right after the news section, because we basically split the
(17:27):
newsletter into two pieces, and the podcast is also in
two pieces. So if you're listening to the non-member version
of the show, then after the news, it's going to
have a little message there. Uh, but if you sign up,
become a member, you know, all that stuff. You basically
get access to the member feed, and the member feed
(17:47):
is full. It's a full episode. Um, with all the stuff,
including this discovery section, which is just absolutely insane. Um,
because I've been doing all these upgrades to my sources,
I've added like a thousand 1500 additional sources, like which
(18:09):
I just put like another ten hours in, like expanding
my opml file and like all my different stuff. And
that's why the quality of discovery has just gone up massively. Um,
and that's why I decided, hey, you know, we really
should be offering more of that benefit to the members
and not just giving it away flat to non-paying members
(18:30):
as well. So that separation is why we're doing the
two different versions. So you might have noticed you're getting
like the standard episode of the newsletter or the standard
episode of the podcast. It's all part of that. All right.
So highly recommended. successor essay to my Sspca article from 2023.
This is basically one of the final forms of what
(18:52):
we can actually do with AI, and I'm going to
do a separate video on this. So I don't want
to go into a whole thing about it. All right. Cybersecurity.
Phenomenal analysis of the cybersecurity market from my buddy Mike
Privett on return on security I call him the Nate
Silver of Cybersecurity Market Analysis. He basically does. He looks
at all the startups. He looks at all the spending,
(19:12):
all the funding, what VCs are doing. It is absolutely
worth a sign up. His podcast or his newsletter is
Return on Security. And main takeaway here is cyber investments
are actually getting back kind of to normal in 2014
or 2024, and they now have over $14 billion in funding.
(19:35):
But I and private equity are playing a much bigger role.
So that makes sense. Massive leak of Black Basta ransomware gangs.
Internal Chats has researchers working to translate and analyze over
500,000 Russian messages. So they're trying to figure out what
all they were saying, but they have to do translation.
(19:56):
Russian hackers are successfully compromising encrypted signal messages from Ukrainian military,
and I think the attack is actually much larger than
that scope. But, um, the whole trick is to get
them to scan malicious QR codes and basically bypassing a
bunch of protections that way. Apple dropped advanced data protection
(20:17):
in the UK so that advanced data protection is basically
end to end protection. The UK said, I want a
backdoor in that and Apple says you can't have a backdoor,
so we'll just turn off end to end encryption for you.
What is the UK doing? Why do they hate themselves
so much? Just massive self sabotage. You could trick Gpts
Chatgpt's operator feature into leaking private user data through prompt injection.
(20:42):
Australia is joining the US in banning Kaspersky products. Not
sure what took him so long. Some researchers found they
could consistently break prompt defenses by feeding models bizarre Indiana
Jones themed adventure stories. Yep yep yep yep. New phishing
as a service platform called Dracula V3 has emerged that
(21:04):
lets criminals clone any brand's website in under ten minutes.
Really good for phishing data leak from top Chinese cybersecurity
company reveals they're offering censorship as a service to help
monitor and control public opinion in China. Topcic. Thanks, Topcic.
OpenAI just banned a bunch of accounts using ChatGPT to
(21:24):
help create a China surveillance tool for tracking anti-China protests
in the West. Yeah, so Chinese surveillance company using ChatGPT
to create a surveillance tool. Awesome. Good that, uh, OpenAI
is watching that stuff and blocking it. And the head
of Australia's intelligence agency is saying multiple foreign states have
(21:47):
been plotting to murder dissidents on Australian soil, and this
is under national security. All right, I anthropic finally dropped
their latest model, uh, sonnet 3.7. And there's actually mixed
feedback on this. People are saying it's really smart, really opinionated,
which they like, but it turns out it's actually kind
of ignoring a lot of instructions. There's a lot of
(22:09):
people saying this. I haven't personally experienced it yet. It's
been solid for me so far. But enough people are
complaining about this that I do think it's probably actually real,
and people are getting pretty upset about it, and they're
actually switching back to sonnet 3.5. So I anticipate that
over the weekend or sometime next week, they're probably going
(22:30):
to do like a small dot release on top of this,
or maybe like bump it to who knows, they'll just
do like sonnet 3.7 v2, because they're not very good
at naming, just like OpenAI, or at least people have
been struggling with the naming. Ideally they would call it 3.7.1.
Why not do that? Please do that. Please call it 3.7.1.
(22:54):
That would be the smartest choice, honestly. But anyway, I
anticipate that they fix this like within a week because
enough people are complaining about it. Uh, yeah. Benchmarks look
completely insane. Um, I anticipate that once this gets fixed,
people are going to lock in on 3.7 as being
the thing. It also is capable of thinking. Okay, so
(23:18):
the other thing they released in this model, in addition
to it just being a better, smarter one, is you
can send a thinking, uh, parameter to the API and,
and you give it the amount of time, the amount
of tokens, basically the amount of money and the amount
of effort you wanted to think about a particular thing.
(23:38):
So it's really cool that they integrated that right into
the model. It's basically a thinking model or like a
regular model. Um, and it's just a matter of like
what you send to the API. Um, I have been
mostly using sonnet 3.5. I tend to use Gemini Flash
because it has 2 million tokens. So if something is
like a total monster, like an entire book or something, um,
(24:01):
I usually send that to Gemini Flash because of the
2 million tokens, and it's really good with haystack performance,
but I imagine that my go to will start to be, um,
sonnet 3.7 once they fix it and then uh, oh
three from, um, OpenAI or whatever they call the next
(24:22):
version of it, because they're getting ready to start unifying
their names. And they also released this thing called Cloud Code,
which is essentially like, you can think of it as cursor,
except for it's inside the terminal. And a lot of
people are saying they absolutely love that Google is getting
rid of SMS two for codes for Gmail in favor
of QR codes. I was wrong about QR codes. I
(24:42):
want to say that very clearly. I was wrong about
QR codes when they first came out. Whatever. However many
years that was, I was thinking that they were stupid
and fatty. I assumed they were going to be very insecure.
It's actually one of the things in my file that
I've marked as a mistake, because I'm interested, very interested
(25:04):
in understanding what flawed thinking made me think that wouldn't
work because the smarter the AI gets, it can actually
look at that. And most importantly, look at all my
current thinking and maybe find flaws in my current thinking
based on finding the root kernel of my mistake I
made with the QR code. So that's why I think
(25:28):
this is fascinating. Nathan Young wrote a wonderful letter to
future artificial general intelligence about the importance of consciousness this
thing is insanely awesome. Look at this thing. There is something.
Look at this. There is something called consciousness. And it's
the most important thing in the universe. I don't know
(25:49):
if you the eye analyzing this are conscious, but I
hope you are. I don't know what state your universe
is in or what place human consciousness holds in that world,
if any. The main thing I want to say if
you should stumble upon this piece is this. There really
is something called consciousness and it is worth being a
(26:09):
top concern. Isn't that great? It's a letter to I
to future I, I like it, I like it a lot.
Humane AI pin fell from the sky and hit the ground.
I was signed up. I don't even remember if I
gave them money. I've signed up for so many devices
and services and a lot of them didn't work out. But.
And by the way, how was our one still thing?
(26:31):
The biggest flop ever when it came out and it's
still around and kicking. Then you have the humane AI one,
which seemed like it was going to go well and
now it's folded, so can't really predict these things. Elon
has been talking non-stop about how grok three isn't filtered,
and it's super smart, and how XYZ mission is to
(26:51):
pursue the truth no matter what. Great goals, which I
absolutely support. But tons of people are pointing out that
he's starting to filter or censor results that are critical
of him, and like, he can't actually have this both ways. Like,
either grok is smart or he's being called out for
a good reason. And, um, yeah, it's funny. If you
(27:11):
ask grok, I don't know if this is still true
right now, but if you ask grok, like, who is
the biggest source of misinformation right now, having the worst impact,
it would come back and say, actually, Elon or this
administration or something like that. So I think he's going
to basically say, well, that's because, uh, you know, it
(27:35):
was trained on, you know, buy us stuff. But I
don't think he could use that for for too long. Technology.
Software engineering job listings have fallen to a five year low,
with indeed postings at just 65% of 2020 levels. The
reason January 2020 is important because it's pre-COVID only 65%,
(27:55):
so 35% lower than pre-COVID. That's the important piece here.
Interesting analysis of how PMS and engineers are merging because
of AI. And my analysis here is it's basically knowing
what you want to build, knowing why you want to
build that instead of something else, and then pursuing it
and actually doing so. Apple is putting half $1 trillion
(28:17):
into US tech manufacturing with a huge focus on AI
and chip production. Meta's Ray-Ban glasses, crushing it with 2
million units, talked about that. YouTube is officially beaten Spotify
and Apple as the top source for podcasts. This is
what I do for when I watch TV, I am
watching podcasts, essentially, and sometimes they're like produced. Sometimes it's
(28:40):
like exactly like this where I'm just talking on the camera,
going through stories. But, um, Matthew Berman is one of
my favorites. Uh, fireship is one of my favorites and
a whole bunch of AI builders. I just watch their stuff. Um,
I don't care if it's produced. Not produced as long
as the content is good. Superhuman just announced a major
(29:02):
AI focused release that integrates AI super deeply into your
email workflows. I wish I could pull up my email
right now and drag it over here, but that is
too risky. I now am using. I maxed out the
number of auto, um, labels that I turned on. I
turned on auto labels. This thing is now filtering my email. Um,
(29:24):
it already has access to my email because it's my client.
So when the emails come in, it's auto filtering. According
to these rules, I set up probably 10 or 20.
I set up. So it's like, is this a pitch?
Is it a pitch about coming on a podcast? Is
it a media person trying to get me to make
a comment about something? It is, um, somebody tried to
(29:46):
come on my podcast. Is somebody trying to get me
on their podcast? Is it feedback from the community? Is
it somebody wanting to collaborate on building something and they're
all colored tags? Um, also things related to like fabric
and other open source projects that I'm working on and
my eyes are getting pretty good at like filtering that
(30:06):
pretty quickly. You could also separate them into separate, uh, inboxes,
email inboxes as well. I think superhuman is like 30
bucks a month, uh, which is a lot of people
think is a lot to pay for email. I would
challenge you on that. If you are better at email,
even by a tiny fraction of a percent, that's probably
(30:29):
worth way more than $30 a month. That's the way
I think about like, all service fees, but, um, yeah,
I just want to mention that it is actually a
paid service and no, I'm not sponsored. Otherwise I would
have mentioned that it would have been a sponsored section,
or at least I would have called that out. Alibaba's
CEO Eddie Wu says they're going all in on AGI development.
(30:50):
Join the club. Humans. New research says despite saying intelligence
matters more, both women and their parents choose the more
attractive guy if she has, like two options or multiple options.
Tech executives are now attending psychedelic slumber parties, where they
use ketamine therapy to reset their minds and escape mental ruts.
(31:14):
That's why I have the invite feature for my new
email client. Uh, I've not heard about these parties. Why
am I not invited? I'm offended. I'm not saying I'm
going to go. Not saying I'm not going to go either.
Gallup says LGBTQ plus identification in the US is now 9.3%,
(31:34):
which is nearly triple what it was in 2012 when
Gallup started tracking this. So yeah, like triple. And what
is that, 12, 13 years? 13 years? Never do arithmetic
on camera. Ellen's now asking federal workers to list what
they did last week or get fired, which like many
things with him, has me cheering and wincing. Love the
(31:58):
efficiency push. And I think that's how he's able to innovate.
But my problem is he's not building something from scratch here, okay.
And then firing people who are not innovative enough at
building things from scratch, which he's doing with like, rocket
companies or Optimus or something like that. In this case,
we have people who actually rely on these services. Right.
(32:20):
And it's hard to tell when when you're going in
there and you're just like canceling programs and like canceling
credit cards and killing off money. A lot of that
money that's going out could be doing really important things.
And it's the same with these employees. You might have
somebody who's doing extraordinarily good work, and they're actually helping
save the lives of multiple people, you know, thousands or
(32:43):
millions of people across the country, but they're really bad
at responding to email or they're really bad at explaining themselves.
Maybe they're super humble, maybe they're bad at writing, but
or maybe they're, like, intimidated by talking to bosses, or
maybe they're just traumatized by the whole thing and they
can't respond, well, that person gets fired, potentially. I don't
(33:06):
know if that's actually going to happen, but let's say
that person gets fired. That is a lot of collateral
damage for a thing that I, I think he is
trying to do good and might actually produce good. But
how do you calculate the collateral damage of of what
could result from this compared to the potential good? That is,
(33:28):
that is a calculus that people massively need to be
thinking about. Bureau of prisons is moving forward with plans
to house trans inmates based on birth, sex rather than
gender identity. Yeah, I've got a lot to say about
this one. Um, there are trans women who have been
on drugs for ten, 15, 20 years, and that is
(33:51):
how they maintain their, their, uh, their gender, their identity,
their health hormone levels, how they appear outside to the world. Um, and,
you know, have good, decent lives, you know, functioning in
that way. Now, in this case, we're talking about prison.
So there are some other variables going on here. But
(34:13):
we are not allowed to torture people in prison. We
are not allowed to starve people in prison. We must
give them X amount of time outside. There are rules
based on human morality and human decency for how you
treat inmates. And that's for people who are already in
(34:33):
jail for committing some sort of crime. So it's not
like we're saying, um, they haven't done anything. We we
understand that they've done something. All these rules apply also
to offenders. The worst offenders, absolute murderers, like, uh, child
related crimes. They, no matter who it is, cruel and
(34:55):
unusual punishment is still illegal. This, I would argue, can
classify in some cases as cruel and unusual punishment. Maybe
even most of the cases. I don't know the numbers here.
I don't know the like divides. I don't have data
on that, but I can pretty much guarantee you that
there are at least some cases like this where this
(35:17):
will absolutely be to the future. Looking back at us,
this will be considered cruel and unusual punishment. And that's
super fucked up. Heart doctor explains how swollen fingertips, leg
edema and changes in eye color can predict an impending
heart attack. My cardiologist. Um, not mine, but my cardiologist
(35:40):
buddy says it's important to know that just because you
don't have these particular signs doesn't mean you're okay. So
something to keep in mind. 27 year old woman's viral
post about girlhood FOMO reveals a widespread loneliness crisis among
women in their 20s and 30s missing close female relationships.
(36:00):
Everyone's missing missing close relationships. That's why I'm always pinging
my homies. Taylor Swift lost 144,000 Instagram followers after getting
booed at the Super Bowl, 144,000. Her boyfriend actually gained followers.
Look at Edward Abbey's raw, honest writings about how to
live fully and die on your own terms. Really good
(36:22):
piece there. Read the whole thing. Neuroscientists argues that extremely
high IQs are basically fictional and even Einstein probably had
around 120 or 130. I have always believed this. Well,
I mean, ever since, like reading a bunch of books about, uh,
evolutionary biology and psychology and everything, basically a whole bunch
(36:43):
of Robert Sapolsky stuff, I think it's likely that IQ
tops out, um, especially in value somewhere like this person,
this neuroscientist is saying around like one 2130. And then
when you start going off into like much higher scores,
first of all, the scores kind of just don't mean anything.
(37:04):
The error rate is massive at those levels, but really
it becomes more about like specialization. It becomes more like,
you know, um, dropping the, uh, the toothpicks and instantly
being able to count them. It's more like very specific,
like parlor tricks. Um, and it seems like the g
(37:25):
loading doesn't scale and give you like the expanded benefits
all the way up into like one, 31 4151, 60.
So it's almost like those numbers don't really exist or
don't really matter. I think what the more important thing
is is that below 100, it can be seriously problematic.
(37:48):
Above 100 starts to get you a lot of benefits.
But really, like most smart people, I think that we
know are probably like really smart people that you know
and think of are probably in the one 2130 range.
So I think that's fascinating because it's actually not that
(38:09):
number that matters. What matters is what you stack on
top of it. Most importantly, I would argue, is creativity.
Then like, um, determination, um, drive, ambition, grit. So you
start stacking those together on top of like 120 IQ,
then you're doing something because there's tons of people with
(38:30):
like very high IQs not motivated, not creative, not empathetic,
not like emotionally intelligent. What are they doing? What are
they offering to the world? And I would argue often,
not very much. NASA contracted lunar lander just beamed back
some gorgeous shots of the moon ideas. Blogging might get
even more important. So what if blog like content from
(38:54):
actual real people with thoughts? Because some of the highest
rated signal for AI. So we know we have this
data crisis with AI, right? Well, maybe bloggers are going
to be the main source. Um, I would say writers
in general, but bloggers as basically writers, and of course
you have like high quality people at some magazines and newspapers.
(39:17):
But how how many of those people are left? They're
mostly going to Substack. Right. So now Substack, Beehive personal blogs.
That is where the actual new ideas are coming out onto.
And I will say YouTube because YouTube is basically blogging.
I mean, it's all the same shit. YouTube is basically blogging.
(39:41):
It's just video blogging. It's all the same. The point
is YouTube. Substack. Beehive. Personal sites RSS. This becomes the
new source of actual novel and creative content, so I
might start valuing it very, very highly. All right. This
(40:01):
is the end of the standard edition of the podcast,
which includes just the news items for the week to
get the rest of the episode, which includes my analysis,
the discovery section that has all of the coolest tools
and articles. I found this week, the recommendation of the
week and the aphorism of the week. Please consider becoming
a member. As a member, you'll get lots of different
(40:23):
things from access to our extraordinary community of over a
thousand brilliant and kind people in industries like cybersecurity, AI,
technology and the humanities. You also get access to the
UL Book Club, dedicated member content and events, and lots,
lots more. Plus, you'll get a dedicated podcast feed you
(40:44):
can put in your client that gets you the full
member edition of this podcast. To become a member and
get all of that, just head over to Daniel Comm
Slash upgrade. That's Daniel miessler.com/upgrade. We'll see you next time.
Unsupervised learning is produced on Hindenburg Pro using an SM
(41:04):
seven B microphone. A video version of the podcast is
available on the Unsupervised Learning YouTube channel, and the text
version with full links and notes is available at Daniel
Mysa.com slash newsletter. We'll see you next time.
Does your app get fake sign ups, throw away emails
or users abusing your free tier? Or worse, bot attacks
and brute force attempts? Work OS radar can block all
of this and more. Simple API gives you advanced device
fingerprinting that can detect bad actors, bots, and suspicious behavior.
Your users trust you. Let's keep it that way. Check
(00:23):
out work OS Radar at work OS comm slash radar.
That's work os.com/radar. Unsupervised learning is a podcast about. Trends
and ideas in cybersecurity, national security, AI, technology and society
and how best to upgrade ourselves to be ready for
what's coming. All right. Welcome to unsupervised Learning. This is Daniel,
(00:51):
episode 470. Let's get into it. Reading a couple of
new books in addition to this month's books for UL.
So the UL books are 1984 by Orwell and Animal
Farm by Orwell. And these two books I just added
to the list. I'm about done with the Technical Republic
(01:11):
and about to jump into money, lies and God. And actually,
there's another one I'm going to add to this list,
somebody who came on the Goodfellas podcast. So I'm going
to go find his book as well. But, um, yeah,
having a good time with some reading. Got a friend
in Guatemala who was recently laid off senior engineer, focused
(01:33):
around monitoring solutions, but really can do lots of different stuff.
So I recommend you check out his LinkedIn and see
if if you might have anything. My friend Monica is
offering 25% off her Security Leadership Masterclass, which I consulted for,
and I think it's really good for people trying to
get into leadership around security. I got a funny meme
(01:56):
here about, um, how to calm down. A buddy who
loses a bunch of money in crypto. It's just. It's
just funny. Nothing useful there except for laughter, which is useful.
LinkedIn posts about my ultimate app, which I keep iterating on.
I'm going to go ahead and click through to this one.
So this is essentially like the main app that I've
(02:19):
been building for the longest time. And I've got like
different versions of it. I've got like a commercial version,
I've got the overall arching one, but essentially it takes
inputs from any medium, from any platform and does a
ton of analysis to determine how good it is. Okay,
so just imagine that this is like the most important
thing to me is most important like current buildable use
(02:44):
for AI. So all this stuff is happening in the world, right.
All this news, new intelligence is is dropping open source intelligence,
recon vulnerabilities are popping up on websites. Um, my favorite
thinkers are releasing a new blog, blog post, a new essay,
a new article, a new video on YouTube comes out right.
(03:06):
All the authors that I follow, they're releasing new books, right? Um,
comments are being made which might reveal something really, really interesting. Uh,
new art is coming out, new books, new movies that
I should be watching. Um, comments from my friends, new
companies being launched. All this stuff is happening, you know,
(03:28):
millions of things per day. Hundreds of thousands, tens of
thousands or whatever, uh, pieces of content coming into my
RSS feeder. And all of that is, is something I
could potentially benefit from. Okay. So this app that I've
been working on for the longest time, really starting, uh,
(03:49):
beginning of 23, is essentially the consumption mechanism for this.
Pull it all together, then use AI to run it
through a filter. Okay. That filter. And I've already built
this way back then. The filter basically does a quality
check on all of these things to see if it's
(04:09):
full of interesting ideas, if it's creative, if it's novel,
if it's surprising, and all these various different aspects similar
to the label and rate, um, fabric pattern that I
have out there that's public. This is just like a
much more advanced version of it, with a lot more
labels and stuff like that, because based on that, the
(04:32):
AI can then or a different AI or a different
part of the same AI can then sort what to
do with it, right? Because it might be an alert.
It might send me an alert if something is important enough, like, hey,
a new vulnerability was found on one of your websites,
or a new vulnerability was found on a bounty program
that I'm monitoring. And by the way, the AI is
(04:56):
telling me my AI is like, oh, by the way,
I found the vulnerability. I found how you could actually
exploit it. I wrote that up into a report and
I submitted it to the thing. So we are waiting
to see if we're going to get paid from this bounty. Right.
So that that's a cool thing. And it might send
me an email which tags it with a certain thing
that says awaiting for bounty payout or something like that.
(05:19):
Or it might be that I just want to know if, um,
this particular company buys any new companies, or I want
to know if this particular company does anything to, like,
raise money. All of these could be like alerts that
I have set up. But it starts with the consumption
of the world and the processing through this AI. Now
(05:42):
I've already built this I've got is probably a version
like three or something, depending on how you call the versions.
But what I want to mention about this is that
the real version of this, the real version of this,
which I've talked about in my other videos, especially the
predictive path AI. Video is that the real version of
(06:04):
this is my AI buddy, my AI assistant, my AI friend, my.
AI my digital assistant. Okay, my universal primary digital assistant
is going to be in charge of doing this for me. Okay.
My APIs that are available to me are the ones
that are doing this processing right. These agents, these automation workflows,
(06:28):
these things that are just running on my behalf. They're
out there in the world collecting, analyzing, assessing, um, labeling,
doing all those things for me. But the quarterback for
all of this is my Da. My Da is monitoring
how many feeds I have, if they're the right feeds,
if this feed like we thought it was awesome, but
(06:49):
it's actually garbage. It might have the permission to go
in and clean that up and fix it. Right. So
the future of this is I don't really ask any
questions about this. I'm not like typing very much. I'm not, um,
really manually interacting with this too much at all. Because
if I want a new source, I say, hey, add
(07:10):
this to my source, okay? While I'm looking at a
page on the screen, hey, add this to my our
list of sources in my says in my ear, yeah,
no problem. Just added it. Okay, I'm watching a new
YouTube video, for example. I'm like, yeah, add this to
our sources and it's like it runs the API command
to go add that. It's now part of the workflow.
(07:33):
It's now in the system right now. How could it
do that? Well, it's going to do that because it's
watching all my screens or it's watching with the camera.
However it's doing it. There's going to be lots of
different ways. But the point is, my Da will be
part of my life. Um, Apple is talking about having
cameras on its AirPods. I don't know if that will
be the next version. Probably not. It'll be the version
(07:55):
like 2 or 3 versions after the current version, but
I'm wearing them now. Like if if these were like
over here and in the back, it had a camera
that was pointing behind me. Amazing. Obviously it should have
a camera that's pointing forward, pointing forward, pointing behind me
with full access to my life context. Full access to
(08:17):
my screen. It should also have access within the screen
as well, so we could look at processes and stuff
like that. But let's just say you could only just
see the screen. I could still ask it things and
it could still do lookups. It could still summarize what
I'm looking at, for example. So this is really the interaction.
The interaction becomes me just talking to my AI and
(08:39):
my AI giving me back results. Now the second level
of this, which comes a little bit later because the
technology just is not quite there yet, but it's more
like the meta glasses. Okay, the meta Metaclasses. As you
saw in the newsletter last week, or maybe it's this
week and we're about to talk about it. The Metaclasses
(09:00):
have sold 2 million units. They are the direction that
things are going. I imagine Apple is going to move
in this direction as well, and kind of offload a
bunch of the processing to the phone, but the next
version of this app that I'm building, the reason I'm
building all these services that are working for me, coming
through my digital assistant, which also doesn't fully exist yet. Okay,
(09:24):
that's also a place everyone is pushing into, including Apple.
But the real thing is that I am watching. I
am talking to somebody and they're like, hey, you know,
you should get in on this deal and blah blah, blah.
And it's crypto related and it's a meme coin. And
I'm just like, uh, well, not interested. Don't want to
(09:45):
hear about that. Um, but let's say I was interested
and let's say it did sound good. In the meantime,
my Da is calling all my APIs. Okay, I have
an API for researching someone. I have an API for
creating a dossier on somebody. I have an API for monitoring,
watching someone's face and determining characteristics of lying. So while
(10:10):
I'm looking with my apple glasses, which is just like
they look like regular glasses, this is pretty far in
the future. Like five years maybe, maybe shorter, maybe like
three years with meta or something. Who knows? But let's
just say three to 5 to 10 years. Okay. As
you're watching, I have this little dial in the corner
of my glasses and it's like, it's like a bullshit meter, right?
(10:33):
It's like, yeah, no, don't believe that. And it could
even be printing out little thing in the text or whatever.
That's like that claim he just made is incorrect. That
was kind of bullshit. Oh, that was kind of smart
or whatever. You could do whatever. You could have an
outline around them based on the type of person you're
looking at. I talked about this in that blog post,
(10:54):
which you should go check out. Actually, here's the video
for it. So you've got to think about what is
possible from your Da when your Da can also control
what you are seeing. The overlay on top of the world. Okay,
so when I walk into a Starbucks and I'm looking
for a new partner, I'm looking for a girlfriend, which
(11:16):
I'm not, by the way, but let's say I were, um,
there could be different outlines based on what, um, the
girls in the Starbucks are broadcasting in their demons. They're
broadcasting that they're creative. They're broadcasting that they like programming,
they're broadcasting that they want to start a business. They're
broadcasting whatever. And my Da could then place different outlines
(11:41):
around them based on highlighting that for me, knowing what
I'm looking for. Right. And more importantly, there's going to
be 1000 or 1 million companies out there pitching to
my Da that their filter on reality is the best one,
because they have the coolest outlines, and they have the
coolest little animations that go around people who are like
(12:02):
artists or engineers or whatever. So this is kind of
these are the steps. This is what I'm building. Like
the stuff I put in that book from 2016, the
stuff I've been talking about in these posts, like I'm
not waiting for it to happen, I'm building the pieces
right now. Like I can't build the hardware. Like I'll
never mess with that. I have to wait for that.
(12:24):
But I can maybe build the assistant, or at least
build something to go on top of the assistant. And
I can absolutely build these services. And this is like
the most important service. This is the gathering and filtering service.
So that is what that's what I've made. Oh and
by the way I the commercial version of this that
(12:44):
I already put out, let me just click this and
see like what we get here. This is my actual threshold.
This is my personal feed for threshold, and I've actually
never showed this. Have I never showed this? I don't
think I've ever showed this before. Let me show you
what this is like. So this is my threshold. And
look what I can do in here. I have a
(13:07):
score here that I can, uh, set the minimum threshold
of quality that has to. The content has to be.
So I have thousands of things like. This is not theory.
I've already built this. Thousands of things are coming into
this thing being parsed, being filtered, being labeled, being rated. Right.
(13:28):
Watch this. It's at 85. I slide this down to 60.
Look I have all these categories as well I hit save.
Look what it does over here. What it just readjusted.
It just readjusted. And watch this watch this view analysis
I have summaries of the contents. You could decide if
you actually want to go and read it, or if
(13:49):
you actually want to go and watch the video and
look at this surface level, middle level, deep level, hidden
message analysis, the ideas that are actually in here, the
recommendations that came out of the content. This thing helps
people save time, going from like hundreds of thousands of
feeds down to a small number of feeds. And even
(14:13):
when you get the small number of feeds that pop
up in this list right here. So let me just
take this to like 95 save. So even when this
thing pops up, I could still decide if I want
to read it or not based on this I usually
keep it around middle. So three ideas, a little review
of it, recommendations coming out of it. It's insane. And
(14:36):
then I always have the link to go click it
and read the actual full thing. So this thing saves
me like infinite time. And I find so much content
that I would never find before. And here, here's the
ultimate idea for this. And this is the reason I'm
building this entire system separate from this app here in
the commercial app. I don't care if the best idea
(14:58):
came from like a 13 year old girl in Nairobi,
because her idea might be as good or better than
Marc Andreessen's latest idea that just came out. But Marc
Andreessen is going to get all the press. It's going
to be through all my feeds. So if I just
rely on that, it's going to be hype cycle based
(15:19):
on social dynamics. So it's going to be surfaced to me,
but I am blocking that out. My algorithm does not
care about how popular that person is. It doesn't care
how about how popular that it turned into being going viral?
It doesn't care. It's judging the quality of the ideas
(15:40):
and the the novelty of the ideas and the creativity.
That's what's determining whether or not I'm going to see
a thing. So that's like the commercial version of this basically,
which is already out. And of course I'm going to
improve that as well. But it's kind of like not
the main point. The main point is this right here,
(16:00):
my interface to the world. This is why I talk
about augmentation okay. Augmentation is this is this is why
I am so excited about, I think of how much
smarter I can be about the world if I'm being
constantly fed the best ideas from the entire planet, the
(16:21):
more stuff I can find from unknown people. And by
the way, when I when I hear from unknown people,
I give them a shout out. I broadcast them out
on on X or on blue Sky or in the newsletter,
or I reach out to them and ask them if
they want mentoring. Like it is so fun to find
(16:45):
like nascent talent in the world and like, try to
give them confidence and lift them up and kind of
broadcast them out. And you've seen me do it a
million times, and it's just it's unbelievable to connect with
people who have interesting ideas. And that's what I love
about this thing is it finds interesting ideas regardless of source.
(17:06):
Forget the source. I don't care about the source. I
want to talk about ideas. All right, so that was that. Um,
the discovery section in the newsletter is getting absolutely insane. Uh,
there's going to be a cutoff. Um, in this episode, um,
right after the news section, because we basically split the
(17:27):
newsletter into two pieces, and the podcast is also in
two pieces. So if you're listening to the non-member version
of the show, then after the news, it's going to
have a little message there. Uh, but if you sign up,
become a member, you know, all that stuff. You basically
get access to the member feed, and the member feed
(17:47):
is full. It's a full episode. Um, with all the stuff,
including this discovery section, which is just absolutely insane. Um,
because I've been doing all these upgrades to my sources,
I've added like a thousand 1500 additional sources, like which
(18:09):
I just put like another ten hours in, like expanding
my opml file and like all my different stuff. And
that's why the quality of discovery has just gone up massively. Um,
and that's why I decided, hey, you know, we really
should be offering more of that benefit to the members
and not just giving it away flat to non-paying members
(18:30):
as well. So that separation is why we're doing the
two different versions. So you might have noticed you're getting
like the standard episode of the newsletter or the standard
episode of the podcast. It's all part of that. All right.
So highly recommended. successor essay to my Sspca article from 2023.
This is basically one of the final forms of what
(18:52):
we can actually do with AI, and I'm going to
do a separate video on this. So I don't want
to go into a whole thing about it. All right. Cybersecurity.
Phenomenal analysis of the cybersecurity market from my buddy Mike
Privett on return on security I call him the Nate
Silver of Cybersecurity Market Analysis. He basically does. He looks
at all the startups. He looks at all the spending,
(19:12):
all the funding, what VCs are doing. It is absolutely
worth a sign up. His podcast or his newsletter is
Return on Security. And main takeaway here is cyber investments
are actually getting back kind of to normal in 2014
or 2024, and they now have over $14 billion in funding.
(19:35):
But I and private equity are playing a much bigger role.
So that makes sense. Massive leak of Black Basta ransomware gangs.
Internal Chats has researchers working to translate and analyze over
500,000 Russian messages. So they're trying to figure out what
all they were saying, but they have to do translation.
(19:56):
Russian hackers are successfully compromising encrypted signal messages from Ukrainian military,
and I think the attack is actually much larger than
that scope. But, um, the whole trick is to get
them to scan malicious QR codes and basically bypassing a
bunch of protections that way. Apple dropped advanced data protection
(20:17):
in the UK so that advanced data protection is basically
end to end protection. The UK said, I want a
backdoor in that and Apple says you can't have a backdoor,
so we'll just turn off end to end encryption for you.
What is the UK doing? Why do they hate themselves
so much? Just massive self sabotage. You could trick Gpts
Chatgpt's operator feature into leaking private user data through prompt injection.
(20:42):
Australia is joining the US in banning Kaspersky products. Not
sure what took him so long. Some researchers found they
could consistently break prompt defenses by feeding models bizarre Indiana
Jones themed adventure stories. Yep yep yep yep. New phishing
as a service platform called Dracula V3 has emerged that
(21:04):
lets criminals clone any brand's website in under ten minutes.
Really good for phishing data leak from top Chinese cybersecurity
company reveals they're offering censorship as a service to help
monitor and control public opinion in China. Topcic. Thanks, Topcic.
OpenAI just banned a bunch of accounts using ChatGPT to
(21:24):
help create a China surveillance tool for tracking anti-China protests
in the West. Yeah, so Chinese surveillance company using ChatGPT
to create a surveillance tool. Awesome. Good that, uh, OpenAI
is watching that stuff and blocking it. And the head
of Australia's intelligence agency is saying multiple foreign states have
(21:47):
been plotting to murder dissidents on Australian soil, and this
is under national security. All right, I anthropic finally dropped
their latest model, uh, sonnet 3.7. And there's actually mixed
feedback on this. People are saying it's really smart, really opinionated,
which they like, but it turns out it's actually kind
of ignoring a lot of instructions. There's a lot of
(22:09):
people saying this. I haven't personally experienced it yet. It's
been solid for me so far. But enough people are
complaining about this that I do think it's probably actually real,
and people are getting pretty upset about it, and they're
actually switching back to sonnet 3.5. So I anticipate that
over the weekend or sometime next week, they're probably going
(22:30):
to do like a small dot release on top of this,
or maybe like bump it to who knows, they'll just
do like sonnet 3.7 v2, because they're not very good
at naming, just like OpenAI, or at least people have
been struggling with the naming. Ideally they would call it 3.7.1.
Why not do that? Please do that. Please call it 3.7.1.
(22:54):
That would be the smartest choice, honestly. But anyway, I
anticipate that they fix this like within a week because
enough people are complaining about it. Uh, yeah. Benchmarks look
completely insane. Um, I anticipate that once this gets fixed,
people are going to lock in on 3.7 as being
the thing. It also is capable of thinking. Okay, so
(23:18):
the other thing they released in this model, in addition
to it just being a better, smarter one, is you
can send a thinking, uh, parameter to the API and,
and you give it the amount of time, the amount
of tokens, basically the amount of money and the amount
of effort you wanted to think about a particular thing.
(23:38):
So it's really cool that they integrated that right into
the model. It's basically a thinking model or like a
regular model. Um, and it's just a matter of like
what you send to the API. Um, I have been
mostly using sonnet 3.5. I tend to use Gemini Flash
because it has 2 million tokens. So if something is
like a total monster, like an entire book or something, um,
(24:01):
I usually send that to Gemini Flash because of the
2 million tokens, and it's really good with haystack performance,
but I imagine that my go to will start to be, um,
sonnet 3.7 once they fix it and then uh, oh
three from, um, OpenAI or whatever they call the next
(24:22):
version of it, because they're getting ready to start unifying
their names. And they also released this thing called Cloud Code,
which is essentially like, you can think of it as cursor,
except for it's inside the terminal. And a lot of
people are saying they absolutely love that Google is getting
rid of SMS two for codes for Gmail in favor
of QR codes. I was wrong about QR codes. I
(24:42):
want to say that very clearly. I was wrong about
QR codes when they first came out. Whatever. However many
years that was, I was thinking that they were stupid
and fatty. I assumed they were going to be very insecure.
It's actually one of the things in my file that
I've marked as a mistake, because I'm interested, very interested
(25:04):
in understanding what flawed thinking made me think that wouldn't
work because the smarter the AI gets, it can actually
look at that. And most importantly, look at all my
current thinking and maybe find flaws in my current thinking
based on finding the root kernel of my mistake I
made with the QR code. So that's why I think
(25:28):
this is fascinating. Nathan Young wrote a wonderful letter to
future artificial general intelligence about the importance of consciousness this
thing is insanely awesome. Look at this thing. There is something.
Look at this. There is something called consciousness. And it's
the most important thing in the universe. I don't know
(25:49):
if you the eye analyzing this are conscious, but I
hope you are. I don't know what state your universe
is in or what place human consciousness holds in that world,
if any. The main thing I want to say if
you should stumble upon this piece is this. There really
is something called consciousness and it is worth being a
(26:09):
top concern. Isn't that great? It's a letter to I
to future I, I like it, I like it a lot.
Humane AI pin fell from the sky and hit the ground.
I was signed up. I don't even remember if I
gave them money. I've signed up for so many devices
and services and a lot of them didn't work out. But.
And by the way, how was our one still thing?
(26:31):
The biggest flop ever when it came out and it's
still around and kicking. Then you have the humane AI one,
which seemed like it was going to go well and
now it's folded, so can't really predict these things. Elon
has been talking non-stop about how grok three isn't filtered,
and it's super smart, and how XYZ mission is to
(26:51):
pursue the truth no matter what. Great goals, which I
absolutely support. But tons of people are pointing out that
he's starting to filter or censor results that are critical
of him, and like, he can't actually have this both ways. Like,
either grok is smart or he's being called out for
a good reason. And, um, yeah, it's funny. If you
(27:11):
ask grok, I don't know if this is still true
right now, but if you ask grok, like, who is
the biggest source of misinformation right now, having the worst impact,
it would come back and say, actually, Elon or this
administration or something like that. So I think he's going
to basically say, well, that's because, uh, you know, it
(27:35):
was trained on, you know, buy us stuff. But I
don't think he could use that for for too long. Technology.
Software engineering job listings have fallen to a five year low,
with indeed postings at just 65% of 2020 levels. The
reason January 2020 is important because it's pre-COVID only 65%,
(27:55):
so 35% lower than pre-COVID. That's the important piece here.
Interesting analysis of how PMS and engineers are merging because
of AI. And my analysis here is it's basically knowing
what you want to build, knowing why you want to
build that instead of something else, and then pursuing it
and actually doing so. Apple is putting half $1 trillion
(28:17):
into US tech manufacturing with a huge focus on AI
and chip production. Meta's Ray-Ban glasses, crushing it with 2
million units, talked about that. YouTube is officially beaten Spotify
and Apple as the top source for podcasts. This is
what I do for when I watch TV, I am
watching podcasts, essentially, and sometimes they're like produced. Sometimes it's
(28:40):
like exactly like this where I'm just talking on the camera,
going through stories. But, um, Matthew Berman is one of
my favorites. Uh, fireship is one of my favorites and
a whole bunch of AI builders. I just watch their stuff. Um,
I don't care if it's produced. Not produced as long
as the content is good. Superhuman just announced a major
(29:02):
AI focused release that integrates AI super deeply into your
email workflows. I wish I could pull up my email
right now and drag it over here, but that is
too risky. I now am using. I maxed out the
number of auto, um, labels that I turned on. I
turned on auto labels. This thing is now filtering my email. Um,
(29:24):
it already has access to my email because it's my client.
So when the emails come in, it's auto filtering. According
to these rules, I set up probably 10 or 20.
I set up. So it's like, is this a pitch?
Is it a pitch about coming on a podcast? Is
it a media person trying to get me to make
a comment about something? It is, um, somebody tried to
(29:46):
come on my podcast. Is somebody trying to get me
on their podcast? Is it feedback from the community? Is
it somebody wanting to collaborate on building something and they're
all colored tags? Um, also things related to like fabric
and other open source projects that I'm working on and
my eyes are getting pretty good at like filtering that
(30:06):
pretty quickly. You could also separate them into separate, uh, inboxes,
email inboxes as well. I think superhuman is like 30
bucks a month, uh, which is a lot of people
think is a lot to pay for email. I would
challenge you on that. If you are better at email,
even by a tiny fraction of a percent, that's probably
(30:29):
worth way more than $30 a month. That's the way
I think about like, all service fees, but, um, yeah,
I just want to mention that it is actually a
paid service and no, I'm not sponsored. Otherwise I would
have mentioned that it would have been a sponsored section,
or at least I would have called that out. Alibaba's
CEO Eddie Wu says they're going all in on AGI development.
(30:50):
Join the club. Humans. New research says despite saying intelligence
matters more, both women and their parents choose the more
attractive guy if she has, like two options or multiple options.
Tech executives are now attending psychedelic slumber parties, where they
use ketamine therapy to reset their minds and escape mental ruts.
(31:14):
That's why I have the invite feature for my new
email client. Uh, I've not heard about these parties. Why
am I not invited? I'm offended. I'm not saying I'm
going to go. Not saying I'm not going to go either.
Gallup says LGBTQ plus identification in the US is now 9.3%,
(31:34):
which is nearly triple what it was in 2012 when
Gallup started tracking this. So yeah, like triple. And what
is that, 12, 13 years? 13 years? Never do arithmetic
on camera. Ellen's now asking federal workers to list what
they did last week or get fired, which like many
things with him, has me cheering and wincing. Love the
(31:58):
efficiency push. And I think that's how he's able to innovate.
But my problem is he's not building something from scratch here, okay.
And then firing people who are not innovative enough at
building things from scratch, which he's doing with like, rocket
companies or Optimus or something like that. In this case,
we have people who actually rely on these services. Right.
(32:20):
And it's hard to tell when when you're going in
there and you're just like canceling programs and like canceling
credit cards and killing off money. A lot of that
money that's going out could be doing really important things.
And it's the same with these employees. You might have
somebody who's doing extraordinarily good work, and they're actually helping
save the lives of multiple people, you know, thousands or
(32:43):
millions of people across the country, but they're really bad
at responding to email or they're really bad at explaining themselves.
Maybe they're super humble, maybe they're bad at writing, but
or maybe they're, like, intimidated by talking to bosses, or
maybe they're just traumatized by the whole thing and they
can't respond, well, that person gets fired, potentially. I don't
(33:06):
know if that's actually going to happen, but let's say
that person gets fired. That is a lot of collateral
damage for a thing that I, I think he is
trying to do good and might actually produce good. But
how do you calculate the collateral damage of of what
could result from this compared to the potential good? That is,
(33:28):
that is a calculus that people massively need to be
thinking about. Bureau of prisons is moving forward with plans
to house trans inmates based on birth, sex rather than
gender identity. Yeah, I've got a lot to say about
this one. Um, there are trans women who have been
on drugs for ten, 15, 20 years, and that is
(33:51):
how they maintain their, their, uh, their gender, their identity,
their health hormone levels, how they appear outside to the world. Um, and,
you know, have good, decent lives, you know, functioning in
that way. Now, in this case, we're talking about prison.
So there are some other variables going on here. But
(34:13):
we are not allowed to torture people in prison. We
are not allowed to starve people in prison. We must
give them X amount of time outside. There are rules
based on human morality and human decency for how you
treat inmates. And that's for people who are already in
(34:33):
jail for committing some sort of crime. So it's not
like we're saying, um, they haven't done anything. We we
understand that they've done something. All these rules apply also
to offenders. The worst offenders, absolute murderers, like, uh, child
related crimes. They, no matter who it is, cruel and
(34:55):
unusual punishment is still illegal. This, I would argue, can
classify in some cases as cruel and unusual punishment. Maybe
even most of the cases. I don't know the numbers here.
I don't know the like divides. I don't have data
on that, but I can pretty much guarantee you that
there are at least some cases like this where this
(35:17):
will absolutely be to the future. Looking back at us,
this will be considered cruel and unusual punishment. And that's
super fucked up. Heart doctor explains how swollen fingertips, leg
edema and changes in eye color can predict an impending
heart attack. My cardiologist. Um, not mine, but my cardiologist
(35:40):
buddy says it's important to know that just because you
don't have these particular signs doesn't mean you're okay. So
something to keep in mind. 27 year old woman's viral
post about girlhood FOMO reveals a widespread loneliness crisis among
women in their 20s and 30s missing close female relationships.
(36:00):
Everyone's missing missing close relationships. That's why I'm always pinging
my homies. Taylor Swift lost 144,000 Instagram followers after getting
booed at the Super Bowl, 144,000. Her boyfriend actually gained followers.
Look at Edward Abbey's raw, honest writings about how to
live fully and die on your own terms. Really good
(36:22):
piece there. Read the whole thing. Neuroscientists argues that extremely
high IQs are basically fictional and even Einstein probably had
around 120 or 130. I have always believed this. Well,
I mean, ever since, like reading a bunch of books about, uh,
evolutionary biology and psychology and everything, basically a whole bunch
(36:43):
of Robert Sapolsky stuff, I think it's likely that IQ
tops out, um, especially in value somewhere like this person,
this neuroscientist is saying around like one 2130. And then
when you start going off into like much higher scores,
first of all, the scores kind of just don't mean anything.
(37:04):
The error rate is massive at those levels, but really
it becomes more about like specialization. It becomes more like,
you know, um, dropping the, uh, the toothpicks and instantly
being able to count them. It's more like very specific,
like parlor tricks. Um, and it seems like the g
(37:25):
loading doesn't scale and give you like the expanded benefits
all the way up into like one, 31 4151, 60.
So it's almost like those numbers don't really exist or
don't really matter. I think what the more important thing
is is that below 100, it can be seriously problematic.
(37:48):
Above 100 starts to get you a lot of benefits.
But really, like most smart people, I think that we
know are probably like really smart people that you know
and think of are probably in the one 2130 range.
So I think that's fascinating because it's actually not that
(38:09):
number that matters. What matters is what you stack on
top of it. Most importantly, I would argue, is creativity.
Then like, um, determination, um, drive, ambition, grit. So you
start stacking those together on top of like 120 IQ,
then you're doing something because there's tons of people with
(38:30):
like very high IQs not motivated, not creative, not empathetic,
not like emotionally intelligent. What are they doing? What are
they offering to the world? And I would argue often,
not very much. NASA contracted lunar lander just beamed back
some gorgeous shots of the moon ideas. Blogging might get
even more important. So what if blog like content from
(38:54):
actual real people with thoughts? Because some of the highest
rated signal for AI. So we know we have this
data crisis with AI, right? Well, maybe bloggers are going
to be the main source. Um, I would say writers
in general, but bloggers as basically writers, and of course
you have like high quality people at some magazines and newspapers.
(39:17):
But how how many of those people are left? They're
mostly going to Substack. Right. So now Substack, Beehive personal blogs.
That is where the actual new ideas are coming out onto.
And I will say YouTube because YouTube is basically blogging.
I mean, it's all the same shit. YouTube is basically blogging.
(39:41):
It's just video blogging. It's all the same. The point
is YouTube. Substack. Beehive. Personal sites RSS. This becomes the
new source of actual novel and creative content, so I
might start valuing it very, very highly. All right. This
(40:01):
is the end of the standard edition of the podcast,
which includes just the news items for the week to
get the rest of the episode, which includes my analysis,
the discovery section that has all of the coolest tools
and articles. I found this week, the recommendation of the
week and the aphorism of the week. Please consider becoming
a member. As a member, you'll get lots of different
(40:23):
things from access to our extraordinary community of over a
thousand brilliant and kind people in industries like cybersecurity, AI,
technology and the humanities. You also get access to the
UL Book Club, dedicated member content and events, and lots,
lots more. Plus, you'll get a dedicated podcast feed you
(40:44):
can put in your client that gets you the full
member edition of this podcast. To become a member and
get all of that, just head over to Daniel Comm
Slash upgrade. That's Daniel miessler.com/upgrade. We'll see you next time.
Unsupervised learning is produced on Hindenburg Pro using an SM
(41:04):
seven B microphone. A video version of the podcast is
available on the Unsupervised Learning YouTube channel, and the text
version with full links and notes is available at Daniel
Mysa.com slash newsletter. We'll see you next time.
Popular Podcasts
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
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.