SECURITY |
AI |
PURPOSE
UNSUPERVISED LEARNING
is a newsletter about upgrading to thrive in a world full of AI.
It’s original ideas, analysis, mental models, frameworks, and tooling to
prepare you for the world that’s coming.
TOC
NOTES
Hope you’re having a good start to the week!
-
Thomas Roccia
created a web GUI for Fabric called Fabric UI! Really slick!
FABRIC UI -
In the last few days I’ve probably spent 5 hours deep cleaning my mobile
phone environment. Deleted probably 40 applications, phone screen
cleanup, widgets refactor, watch faces cleanup, Focus modes redo, app
subscriptions cleaning, etc.
So much cruft removed. Super cathartic. Timed with new iPhone,
which happens every year, but this year was a major clean.
MY WORK
o1 Use Cases
I did an analysis of
OpenAI’s long-form conversation with the o1 (strawberry) team, where they talked about how they built it, how it’s different from
previous models, what they use it for, etc. I also used
o1-preview to pull out a list of their favorite use cases for
o1 where it’s better than previous models.
MORE
|
THE INTERVIEW
|
THEIR USE CASES
The Strawberry Team’s o1 Use Cases
AI and Many Eyes
I wrote this piece for AT&T Business about
how AI transforms real-time monitoring, threat modeling, and security. It’s
something I’ve talked about before, but this is a deeper look at how I see
it all fitting together.
AI Gets Us the Many Eyes We Were Promised
Discover how AI will transform industries through continuous monitoring and
real-time data processing.
www.linkedin.com/pulse/continuous-ai-daniel-miessler-rdqwc/?trackingId=fQHVrDV4T6SGLE1yX9QR%2BQ%3D%3D
I’m excited to be a keynote speaker at
Swiss Cyber Storm, Switzerland’s premier security conference, held at the stunning Kursaal
in Bern. This year’s theme, “The AI Revolution,” focuses on the
concepts and trends shaping the security landscape. Use code
UNSUPERVISED-LEARNING for 15% off your
registration.
SECURITY
Israel launched an extraordinary attack on Hezbollah using a combination of
supply chain and remote triggering techniques. For the pagers, the Mossad
actually just started a pager company and Hezbollah become one of their
customers. Which allowed them to put explosives in the devices before
shipping them out. Thousands have been severely injured, and something like
dozens or hundreds have been killed (the numbers aren’t clear).
MORE
?I think the biggest thing to discuss in this story is precedent. What
does it mean for the world when a foreign intelligence service starts
blowing up pagers and walkie-talkies of combatants wherever they happen to
be at the time?
In general, because of the extraordinary circumstances of the targets
being Hezbollah operatives, I think it was a surgical and well-thought-out
attack. The vast majority of people even wearing the pagers weren’t
killed, so it’s not like a car bomb where many noncombatants would die as
well. There appear to be very few casualties that weren’t Hezbollah, in
other words, and I think that is a good thing.
I do think it’s a big deal to wage war within a civilian population,
however. Mentally. Visibly. It changes something in our psyche to know
that the opposing side might attack our civilians, even if just
collaterally. So I think anyone considering themselves “the good guy”
should think very carefully about that.
Hezbollah, however, has already been doing that. Actively. On purpose. In
broad daylight. Like it’s literally what they do. They drop rockets on civilians as a matter of course. By comparison,
what the Mossad did here was infinitely humane (a perverse word to use in
war). I just don’t think “better than Hezbollah” is a great standard for
morality.
As another comparison, I believe this Mossad attack was far more careful
and effective than what’s happening in Gaza. It’s 100% true that Hamas is
using their people as human shields. It’s 100% true that Hamas doesn’t
care about their people. And it’s true that Israel must defeat Hamas. But,
to me, with my current level of knowledge, that arithmetic still doesn’t
sum out to it being ok to kill tens of thousands of non-combatants with no
end in sight. Especially when it’s likely that Israel is making many more terrorists than they’re killing by doing so. I just don’t see
how we’re going to—or should—kill all Hamas in Gaza using this strategy.
It just seems bad humanitarian-wise, but even as a strategy for protecting
Israel. There has to be another way.
Anyway, with these targeted supply chain attacks against operatives, I
think the crossover between only hitting combatants and doing very
localized damage was really extraordinary. And given the fact that
Hezbollah is a dictionary definition of a terrorist, I’d say this is about
as clean of an attack as you can get. And that’s not even counting the
psychological aspect.
Here’s an interesting point on that. Actual terrorism is designed to
produce fear. But fear in the civilian
population. What Mossad did was not designed to target or scare Lebanese
people. It was designed to target and scare Hezbollah. That’s the
difference between war and terror. It also has the advantage of making the
population turn against Hezbollah, because you don’t want a terrorist in
line next to you when you’re buying cantaloupe if that person spent last
night building bombs destined for Israel.
That’s a good thing, I think, as a whole. I would just say we should be
very careful about this calculus and constantly re-evaluate it. It’s
pretty clear when it’s Israel, and it’s Hezbollah, and it’s a tiny little
explosive, but the calculation gets a lot more messy when it’s not
Hezbollah, or ISIS, or a group that’s as clearly evil. Or when the
collateral damage increases (see: Gaza).
A security researcher named xyz3va found a “catastrophic” flaw in the Arc browser that let attackers inject arbitrary code into
users’ sessions using just a user ID. MORE
A hacker named Addka72424 has leaked a massive DB of 3.3 billion unique
email addresses on an underground crime forum, claiming it was a personal
experiment to see how much public data he could gather. It’s a 21.8GB
database, and he claims they’re all unique. MORE
Chinese scientists have figured out how to use Starlink satellite signals
to detect stealth aircraft and drones, which are designed to dodge radar.
They ran an experiment in the South China Sea using a DJI Phantom 4 Pro
drone to simulate a stealth aircraft. MORE
Sponsor
Are genAI tools integrated with your other apps?
Find out in minutes with Nudge Security. Nudge Security discovers all genAI accounts ever created by anyone in your org, as well as the OAuth grants that enable data-sharing across apps.
With Nudge Security, you can:
-
Discover all genAI accounts ever created in your org
-
See all users, authentication methods, and OAuth grants
-
Surface and revoke risky OAuth grants
-
Get alerted of new genAI tools or integrations
-
Vet unfamiliar tools with security profiles for each provider
Start a free 14-day trial and get your free genAI inventory in minutes.
nudgesecurity.com/use-cases/mitigate-ai-risks
Nuclei Templates v10.0.0 is out, and includes new Azure Config Review
templates to help automate cloud misconfiguration reviews. This release
builds on previous AWS and Kubernetes security checks, now offering a
streamlined YAML format for Azure.
MORE
Google is making it easier to use passkeys by allowing them to sync
automatically via Google Password Manager across Chrome on Windows, macOS,
and Linux, with Android support and iOS coming soon.
MORE
GreyNoise has been tracking mysterious “Noise Storms” of spoofed internet
traffic since January 2020, but their origin and purpose remain unknown.
These storms, which include a curious “LOVE” ASCII string in ICMP packets,
are suspected to be covert communications or DDoS coordination signals,
among other possibilities.
MORE
Sponsor
Align with Complex Cybersecurity Compliance Frameworks
This ThreatLocker eBook
was designed to help you discover the strategies and toolsets needed to
align with some of the most popular compliance frameworks. Learn how implementing specific technologies can streamline compliance across multiple standards and enhance your cybersecurity posture.
threatlocker.com/ebooks/the-it-professionals-blueprint-for-compliance
Download Now
Continue reading online to avoid the email cutoff…
AI / TECH
Sam Altman just dropped an essay called
The Intelligence Age
about AI progress, and in it he mentions superintelligence being possible in
“thousands of days”. Keep in mind—not AGI. ASI. And thousands of days is
like 3-5 years, roughly, if you read the tone right.
MORE
South Korea’s Electronics and Telecommunications Research Institute has
unveiled “Dejaview,” an AI system that analyzes CCTV footage to predict and
potentially prevent crimes before they happen. Should be fine.
MORE
?Black Mirror as Business Plan.
Not really, though. It just means Black Mirror was good fiction because
it saw where things were going. Along with things like Minority Report, although that was powers-based not AI-based.
Jony Ive has confirmed he’s building a hardware AI device with OpenAI.
The venture, backed by the Emerson Collective, could see up to $1 billion
in funding and is already making waves with a team that includes ex-Apple
designers Tang Tan and Evans Hankey. This thing might crush the others in
the market since AI devices need the synergy between hardware, software, and aesthetics more than most gadgets. MORE
BlackRock and Microsoft are teaming up with the UAE’s MGX to raise $30
billion for AI infrastructure, aiming to leverage it into $100 billion in
investments. This Global AI Infrastructure Investment Partnership will
focus on building data centers and energy projects, primarily in the US. MORE
A Canadian study has found that an AI tool can reduce unexpected deaths in
hospitals by 26%. The tool works by analyzing patient data to predict and
prevent potential complications before they become critical.
MORE
LinkedIn has quietly opted users into using their data to train generative
AI models, raising privacy concerns. Users can opt out by adjusting settings
under “Data for Generative AI Improvement,” but this won’t undo past data
usage.
MORE
A recent study by Ringover found that 76.5% of recruiters preferred
AI-generated headshots over real ones, but only when they didn’t know the
images were AI-generated.
MORE
The Wall Street Journal reports that AI is leading to fewer job openings in
the tech sector, as companies are automating tasks that used to require
human workers.
MORE
A lot of Amazon employees are upset about the requirement to go back to 5
days in the office in January.
MORE
Apple’s iOS 18 update that just came out has RCS support, which means green
bubble Android users will enjoy features like high-res media, typing
indicators, and read receipts when texting their friends with iPhones.
MORE
Apple’s A16 mobile processors are now being produced in the U.S. at TSMC’s
Arizona facility, marking a significant milestone under the CHIPS for
America Fund. These chips, using the N4P process (an enhanced version of
5nm), are initially being manufactured in small quantities, with production
expected to ramp up significantly by 2025.
MORE
?Absolutely loving this self-reliance / American manufacturing push we’ve
seen lately. I think it’ll be increasingly good for the US as things get
less stable geopolitically, and especially as we head towards possible
conflict with China.
Apple’s iPhone 16 now supports wireless firmware restoration. With iOS 18,
users can restore their iPhone 16 by placing it next to another iOS device,
eliminating the need for a Mac or PC. I’m guessing Android has had the
feature since 2002.
MORE
The Apple Watch’s Remote app now lets you adjust volume with the Digital
Crown, invoke Siri, and even control power and audio settings.
MORE
Facundo Olano shares his journey of
turning blog content into an ebook using the jorge site generator.
The process involved creating a jorge project, using epub boilerplate files,
defining chapter layouts, and syncing posts and images.
MORE
HUMANS
Rick Beato argues that music is getting worse because technological
advancements have made music too easy to produce and consume.
MORE
There’s a new study showing that omega-3 fatty acids can help reduce
symptoms of anxiety and depression in mice. Interesting, but you know the
drill with “studies say”. It’s a continuous wait-and-see approach. MORE
The US Department of Energy is rolling out over $3 billion to fund more
than two dozen battery projects across 14 states, aiming to boost domestic
manufacturing and support climate goals. MORE
Astronomers have discovered the largest black hole jets ever observed, named
Porphyrion, stretching an astonishing 23 million light-years—equivalent to
lining up 140 Milky Way galaxies.
MORE
Voyager 1, the 47-year-old spacecraft that’s been cruising through space
since the late ’70s, just fired up some thrusters it hasn’t used in decades.
This maneuver was necessary to adjust its orientation and keep its antenna
pointed toward Earth. Meanwhile, the asphalt on our roads has to be replaced
like every 45 minutes.
MORE
There’s an interesting piece in the Wall Street Journal about how
pediatricians might have inadvertently sparked the peanut allergy epidemic.
The article suggests that by recommending parents avoid giving peanuts to
their young children, they may have increased the risk of allergies instead
of reducing it.
MORE
Ohio is directly funding private religious schools with taxpayer money. The
state has bypassed traditional voucher programs and is now providing
millions in grants to religious schools, mainly Catholic, for infrastructure
projects like building renovations and playground improvements. I’m all for
more structure in schools, but I think we need to be really careful how we
get that.
MORE
Motus is revolutionizing wildlife tracking by using lightweight radio
transmitters to monitor the movements of small flying animals like birds,
bats, and insects. This international network, managed by Birds Canada, has
tagged nearly 50,000 animals across 400 species since 2014, providing
invaluable data on migration patterns and behaviors.
MORE
DISCOVERY
recaptcha-phish — My buddy John Hammond created a phishing tool
that mimics a reCAPTCHA form, tricking users into executing malicious
commands via copy-paste. The attack involves a fake “Verify you are human”
prompt that instructs users to paste a command into the Windows Run dialog,
exploiting the clipboard.
MORE
rga — Ripgrep on growth hormone, letting you search through
PDFs, E-Books, Office documents, and even compressed files like zip and
tar.gz.
MORE
Nuclei Templates v10.0.0
— This release expands on previous AWS and Kubernetes security checks,
offering a streamlined YAML format for easier management and review of
Azure configurations. MORE
Dune Shell —
A new take on the command-line experience, aiming to bring a cozy,
customizable feel that bash lacks. Unlike its predecessor Atom, Dune is
designed with a standalone interpreter, allowing users to create their own
custom frontends and unique shell experiences. MORE
Damn-Vulnerable-Drone
— A drone hacking simulator built on the ArduPilot/MAVLink architecture,
offering a realistic environment for practicing offensive security
techniques. MORE
Sci-fi Ideas
– Someone compiled a massive CSV file containing every sci-fi idea
imaginable. MORE
Eli Bendersky talks about building LLM-powered applications in Go,
highlighting its strengths in handling REST and RPC protocols, concurrency,
and performance. The post walks through creating a RAG (Retrieval Augmented
Generation) server using Go, showcasing different implementations with tools
like the Gemini API, Weaviate, LangChainGo, and Genkit for Go.
MORE
Assetnote talks about their approach to Recon.
MORE
Paul Graham’s One-pager on How to Start a Startup
MORE
IDEAS
Don’t Call Them LLMs
Probably the biggest idea that’s exploded
in my mind lately is Karpathy’s point about LLMs being poorly named. His
stance is that Transformers are general-purpose compute systems, and that
LLMs are actually sequence predictors. And, crucially,
it doesn’t matter what the stream is. We just happen to be sending
language right now. But really what they do is take input, find patterns,
and make predictions.
This is an INSANE way to think about the last couple of years. And the
next couple.
RECOMMENDATION OF THE WEEK
Start reframing your thinking about AI (and specifically LLMs) away from
“just the next token of text” —> Sequence Prediction —>
Answer Prediction.
As Karpathy talks about, the Transformer architecture works on sequences of
ANYTHING—language was just a natural start. It works on whatever you feed
it.
So the recommendation of the week is to:
update your mental model of AI from specific text predictor to generalized answer predictor.
APHORISM OF THE WEEK
❝
People are strange: they are constantly angered by trivial things, but on a
major matter like totally wasting their lives, they hardly seem to notice.
Charles Bukowski
Become a Member to increase Eudaimonia on planet Earth
0 responses on "UL NO. 451: Altman Says ASI in "Thousands of Days""