UL NO. 430: The Courage to be Disliked

How I use local AI models, MI5 vetting research students, the first AI deepfake racism attack, and more…

Author

Daniel Miessler
April 30, 2024

Unsupervised Learning is about the transition from Human 2.0 to Human 3.0 in order to survive and thrive in a post-AI world. It combines original ideas and analysis to provide not just the most important stories and trends—but why they matter, and how to respond.

TOC

NOTES

Hey there,

Someone asked me last week about the best way to run AI models. So here’s a quick breakdown of the different methods and which I use for what.

  1. I prefer using APIs for any programmatic workflows.

  2. For day-to-day use I recommend Fabric because it’s super flexible, has tons of use-cases (Patterns), and lets you use OpenAI, Anthropic, and local models via Ollama.

  3. For super quick demos of open models, plus some API functionality, I like ollama.com.

  4. For using newer, cutting-edge models, my favorite tool is LMStudio.

Asking Uncensored Llama-3 70B how to hack

  1. The big advantage of LMStudio is that you can use the latest models. Including for example, uncensored local models, like my current favorite—QuantFactory/Meta-Llama-3-70B-Instruct-GGUF.

  2. 🔥 You just type in the model you want, and you don’t even need to go get it from Huggingface. It downloads it for you right in the tool, and you can use it right there in the tool, because it’s also a prompting interface.

Other stuff this week:

  • Just had an insanely great book club. The book was The Courage to be Disliked, and we talked for over 70 minutes about it. So thankful to be part of the UL community and to be able to talk about great books (and tons of other stuff). You should come join us.

  • Signed up for my first full-body MRI.

  • RSA next week! Can’t wait to see people.

Ok, let’s get to it…

MY WORK

I finally (thanks to Paul Graham) found a good reason for college—and especially good colleges. I capture the argument here.

The Value of Elite Colleges is Relationships with Elite People

The difference between elite and state education is minimal, but the connections are worlds apart.

danielmiessler.com/p/value-elite-colleges-relationships-elite-people

SECURITY

MI5 is going to start vetting key researchers at Britain's top universities to shield cutting-edge research from espionage—especially from states like China. MORE

💡The US needs to start doing something like this. And not just at universities, but at private companies as well.

In unrelated news, Germany arrested three suspected Chinese spies for stealing tech secrets right after their Chancellor pressed China publicly about IP theft. MORE

After moving forward with the possible ban on TikTok, the U.S. is now looking at a ban on DJI drones, citing national security risks and the company's partial Chinese state ownership. MORE

A former athletic director was arrested for using AI to mimic a principal's voice, falsely accusing him of racism and antisemitism. The AI-generated audio, falsely attributed to the principal, spread harmful stereotypes about Black and Jewish communities, sparking widespread outrage and investigation. MORE

💡Yet another abuse case for deepfakes: character assassination.

As security people it’ll be fascinating and macabre to watch all the different attack types that come out of this. And then the reactions that follow. Such as the pendulum swinging towards not believing attacks.

So deepfake attacks will—among other things—make it harder for people to believe real victims who bring forth evidence.

Sponsor

🔍Are You Wasting Your Email Security Budget?

When every dollar counts, you want to make sure you make the most of what you get. You (hopefully) get funds for anti-phishing tools, but the threat landscape extends beyond the inbox.

With more sophisticated attack flavors at higher volumes than ever, email security must also encompass insider risk scenarios, account takeover protection, and data loss prevention.

See why Material Security is the preferred choice for organizations looking to protect more areas of their Microsoft 365 or Google Workspace footprint under a unified toolkit… and a single line item in the budget.

Nation-state attackers have been using two zero-days in Cisco firewalls to gain access to government networks. This isn’t the first time ASAs have been found to have high-level backdoors in them, but this attack was from new attackers and used new malware. MORE

US lawmakers gave the ok to extend warrantless surveillance under FISA Section 702, widening the net of who can be forced to spy for the government. MORE

Change Healthcare paid $22 million in Bitcoin to ransomware attackers, but files were already compromised so they’re preparing for additional fallout. MORE

Sponsor

🔍The Cybersecurity Platform for IT Teams

Thousands of security tools. Pushy vendors. Endless product trainings. Sound familiar?

IT teams often struggle to implement a strong security program in their organization because buying and maximizing usage of tools is so hard. Oh, and they have a day job too.

Interested? See how you can get 13 tools in one platform here.

The Air Force is moving forward with Anduril and General Atomics to develop unmanned fighter jets, sidelining giants like Boeing and Lockheed Martin. MORE

⚙️ Detectify's External Attack Surface Management solution is now in the AWS Marketplace, making it easier for customers to secure their stuff. | by Detectify | MORE

TECHNOLOGY

Amazon's robot workforce has expanded to over 750,000, overtaking 100,000 human jobs. This dramatic increase in robots, including advanced models like Sequoia and Digit, underscores Amazon's push for efficiency and speed in its operations. MORE

💡I say again—don’t be surprised. Expect it. Companies will do what makes sense for margins, not people. Their responsibility is to shareholders. Hiring humans is not something companies want to do, and they’ll look for any way to avoid it.

And go rewatch The Animatrix.

After a five-month silence, NASA's Voyager 1 is back to sharing its health updates from interstellar space. How in the hell is this thing still working? So impressive. MORE

Google's Prabhakar Raghavan is pushing for speed, urging the 25,000-strong knowledge and information team to adapt to a "new operating reality" with tighter project timelines. In a shift from Google's historically laid-back culture, Raghavan emphasizes the need for agility in the face of growing competition and regulatory challenges. MORE

💡This is progress, but I don’t think the company can fully shift without a senior leadership change. At this point it’s a matter of optics as much as anything.

Police are now using GPT-4 powered body cams that transform audio into reports. MORE

Apple's OpenELM introduces eight open-source LLMs, pushing the envelope for on-device text generation with models up to 3 billion parameters. MORE

Meta's release of Llama-3 has sparked a potential shift in AI development, offering open-source models that rival proprietary ones in performance and flexibility. With over 800 Llama-3 model variations available on Hugging Face, developers now have unprecedented freedom to customize AI for their specific needs, promising a new era of innovation and cost efficiency. MORE

💡I’m really impressed with how good Llama 3 is, and I think open-source—or smaller companies—might have a major advantage over the big players because they will try more things, faster, and more often.

And as I’ve said before, certain tasks only need to be done so well before they’re good enough, which means it won’t be long before most of those tasks are done locally using open-source models.

Crazy times for sure. In a good way.

🗣️OpenVoice is crushing voice cloning with its ability to mimic tone, emotion, and even cross-lingual speech, all while being free for commercial use. | by myshell-ai | MORE

Snowflake just shipped Arctic LLM, an "enterprise-grade" model that's all about efficiency and enterprise needs. Arctic LLM, which took three months, 1,000 GPUs, and $2 million to train, claims to outperform competitors in coding and SQL generation. MORE

🔧 Simone turns your YouTube videos into blog posts. | by rajtilakjee | MORE

The US is finally getting its first high-speed rail, which will go from LA to Vegas in under two hours. MORE

Elon Musk's latest idea is to turn Tesla cars into a distributed AI compute network, akin to AWS but on wheels. MORE

Tesla's Autopilot and Full Self-Driving features are under scrutiny after being linked to hundreds of crashes and dozens of deaths. The investigation revealed a pattern where drivers, overly reliant on Tesla's tech, failed to stay engaged, leading to 956 crashes, including 29 fatalities since January 2018. MORE

💡I can definitely see this being true, but I wonder how that’s going to compare to normal driver error from things like texting, not paying attention, etc.

My guess is that FSD, Waymo, and human-less driving will be a massive net improvement for safety. With occasional stories that scare the crap out of us.

The question is whether we’ll continue on despite those stories if the data shows it’s safer.

For example — a Tesla driver was arrested for vehicular homicide after his distraction with a phone led to a fatal collision with a motorcyclist while on Autopilot. MORE

Volvo cleverly navigated a trade war to bring an affordable Chinese EV to the U.S. market. MORE

HUMANS

Stanford researchers found AI can predict your political leanings just by scanning your neutral face, which is both fascinating and unsettling. Turns out, liberals tend to have smaller faces and conservatives bigger ones. MORE

This article claims that 30% of kids ages 5-7 are on TikTok. MORE

The FTC has banned nearly all noncompete agreements, which some estimates say could boost wages by $300 billion a year by letting people change jobs freely.

💡Really happy about this. I think it’s going to spawn even more innovation in the AI and broader tech space.

Some have raised the issue of IP theft, but we need to stop confusing and inventing new laws. It’s already illegal to steal, so let’s prosecute for that. We don’t enforce anti-theft laws by making it impossible to leave.

The Supreme Court is set to decide if cities can penalize homelessness. MORE

💡I’m all for clamping down on homelessness, but only if we simultaneously build and fund mental hospitals and drug treatment centers.

I’ve been thinking for years that we need a clear way to deal with the homeless.

  1. If you’re mentally ill, let’s get you help and/or get you housed in an institution in a humane way

  2. If you’re on drugs, let’s get you treated and then check for #1. If not #1, go to #3.

  3. If you’re not one of those, let’s figure out if you want to work. If you do, let’s try to get you on your feet with employment

  4. If you’re not any of those, and you just don’t like society and don’t want to participate in it like everyone else, let’s get you somewhere where you’re not bothering people who are trying to be useful

From the stats I’ve seen, most homeless are dealing with drug issues or mental health issues. Let’s help them. That’s our responsibility as good citizens.

And if they don’t want help, and don’t want to work, the role of the good citizen (us) changes to not letting them disrupt everyone else.

Few things are more exciting to me than using AI for actual science innovation and research. AI is now helping physicists explore the vast possibilities of string theory, potentially bridging the gap between complex dimensions and the particles of our universe. MORE

Liron Gertsman's meticulous planning and luck led to capturing a breathtaking eclipse photo, featuring a frigatebird in flight against the "Diamond Ring" effect. MORE

Margaret Macdonald argues that philosophical theories are more like artful stories than scientific facts. MORE

You Are What You Read, Even If You Don't Always Remember It. MORE

Turns out, the reason you're not landing that job in 2024 might be because of "Ghost Jobs." This video dives into the phenomenon where companies post jobs that don't actually exist or have already been filled. MORE

An argument that it's your personality and creative application strategies that truly land you the job. MORE

IDEAS & ANALYSIS

Continuous Recording
It looks like we’re heading towards a place where everything is recorded. Or at least, where you can expect that someone in a public space is probably recording you.

I like what that does for getting value out of conversations, but oftentimes the value is just having the talk in the first place, and having it with someone you trust.

I worry a lot about the fact that people might be more guarded now when having conversations, just assuming it’s being recorded.

I expect there to be some sort of conversation indicator, or visual indicator on someone’s person that politely displays:

  1. That they are recording.

  2. That they are ok with being recorded.

  3. That they are NOT ok with being recorded.

It’ll be interesting to see how continuous recording and AI parsing by DAs will change how we interact with each other. Another option might be that people just stop caring since there are no more ways to be embarrassed or canceled.

I think that’ll be possible in some places, but there will probably be other parts of society that massively value privacy and worry about embarrassment. Basically, we’ll have lots of different societies that think about things differently.

RECOMMENDATION OF THE WEEK

I highly recommend our book of the week, The Courage to be Disliked. It’s about Adlerian Philosophy/Psychology, which is a branch of philosophy/psychology similar to Freud’s and Jung’s, but that I much prefer.

APHORISM OF THE WEEK

❝

Be who you are and say what you feel, because those who mind don't matter, and those who matter don't mind.

Bernard M. Baruch

Thank you for reading.

UL is a personal and strange combination of security, AI, tech, and lots of content about human meaning and flourishing. And because it’s so diverse, it’s harder for it to go as viral as something more niche.

So—if you know someone weird like us—please share the newsletter with them. 🫶 

Happy to be sharing the planet with you,