UL NO. 427: AI's Predictable Future (Video)

Israeli identity reveal, deepfaked content summaries, Altman/Ive device, wealthy kids, Cowen v. Haidt, and more…

Unsupervised Learning is a security, AI, and meaning-focused newsletter that looks at how best to thrive as humans in a world that’s changing faster than ever. It combines original ideas and analysis to bring you not just what’s happening—but why it matters, and how to respond.



Hey there,

  • Going to keynote a conference in NYC this week! Excited to talk about AI, security, and hope for the post-AI future.

  • Conference season is spinning up. Can’t wait to see people. Things I have on deck. Numerous private panels/talks in the Bay Area, talk in NYC, Day of Shecurity, RSA, LocoMocoSec, EDC, big con in Switzerland, possibly others. And that will be enough for the year. The main goal is to not get Covid or Conflu this year.

  • Had a great time at BSides Milwaukee!

  • The soft launch of Threshold is going splendidly. I’m already getting MASSIVE benefits from using it myself, which is honestly my north star for the project. More on that in future posts/videos.

  • Fabric continues to thrive, and I continue to think about the ultimate agent framework. Right now it’s Crew.ai, but I wonder what the model companies are going to have as a response, e.g., Anthropic incorporating tools right into their prompts this week.

Ok, let’s get to it…


I finally turned my big, 9,000-word AI predictions essay into a full video! It’s like 70 minutes long and includes lots of extra narration and detail outside of the text.

Please pass it on to people who would love the content but who could never get through a 9,000 word essay.


Israel's top spy chief, Yossi Sariel, accidentally revealed his identity through an Amazon book sale linked to his real name. MORE

The CVE and NVD databases are struggling to keep up with the massive number of vulnerabilities being created, leading to gaps and inaccuracies. This will need to get addressed somehow, but I’ve not heard any good suggestions. MORE

In Montreal, criminals are using Apple's AirTags to track and steal cars. MORE



Hey, Daniel here. You know how I talk about AI agents all the time? Well it’s happening.

Dropzone.ai is AI agent technology that takes alerts from your environment and performs autonomous multi-step investigations just like a human analyst.

Once it’s finished, it generates decision-ready reports so you can focus on real threats and 10X your team without adding headcount.

No playbooks, code, or prompts required. It just takes alerts and starts investigating them, and then comes back and writes a report.

It’s so good, I just became an advisor.

Panera Bread's week-long IT downtime incident was a ransomware attack. MORE 

CISA's new High-Risk Communities webpage offers cyber hygiene guides, volunteer support, and discounted tools for organizations under cyber threat. MORE


🔍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.

Israel's military used an AI named Lavender to pinpoint 37,000 potential Hamas targets, which is raising questions about the ethics of AI in warfare. MORE


There’s a rumor that Sam Altman and Jony Ive are building some sort of AI wearable device through a new secret company. MORE

OpenAI released improved ways of fine-tuning models. MORE

📄 A new paper shows that adding more agents to large language models can significantly boost their performance. MORE

📄 This paper explores how AI might be leading us towards a 'knowledge collapse' by oversimplifying complex information. MORE

The U.S. is trying to get South Korea to stop chipmaking tool exports to China. MORE

The US is testing energy storage in heated sand, aiming for 135 MW power output for five days straight. MORE

Oura's rolling out Symptom Radar to give you a nudge when your body's showing signs of strain, but it's not calling it illness detection. For obvious reasons. MORE

Amazon's ditching its cashierless "Just Walk Out" tech in Fresh stores for Dash Carts, finding the futuristic checkout too costly and complex. Disappointing. Too early I guess. MORE


New studies are showing that the wealthy are starting to have more kids than the poor, reversing the previous trend. MORE

NASA's doing a live stream of the eclipse. MORE | SPACEX STREAM

Despite a massive earthquake, TSMC's crucial EUV equipment remained unharmed because they have some wicked building stabilization tech. Production was almost fully restored the same day. MORE

The Israeli military dismissed two senior officers and reprimanded three others for an airstrike that mistakenly killed seven World Central Kitchen volunteers in Gaza. MORE

The UK's exporting workers to fill higher-paying US jobs. MORE

U.S. venture capital investments plummeted to $36.6 billion in Q1 2024, which is the lowest quarterly total since 2017. MORE

💡I had a wonderful conversation with Mike Privette from Return on Security about these VC and overall economic trends, which will be released soon.

McKinsey's offering UK employees nine months of pay to voluntarily leave the company. MORE

Gen-Z is going for trades like welding and plumbing over college and student debt. MORE

Home insurers are now using aerial images to decide who gets dropped from coverage. MORE


Another View of Imposter Syndrome (Click for thread)

Tyler Cowen vs. Jonathan Haidt on Teen Girl Depression
Well, I never thought I’d see Tyler Cowen be wrong about something major. But I think he did it here with Jonathan Haidt. He had Jonathan on the show and massively disagreed with him about his research into teen (especially girl) depression being caused by social media. Worth a watch.

The specific thing I think Tyler is getting wrong is the AI Mediation part, which I have written about extensively. He’s right that we’ll have AI mediation, but I think he’s wrong that it’ll work the way he thinks it will.

He seems to think that AI will stop young girls from browsing the most viral and toxic content on social media, and it’ll just send her the summary instead. Problem solved.

Um, no. That’s how HE will use it. And how I will use it. And how Jonathan will use it. And probably you, too.

But for young people consuming viral and toxic content, the content itself is the point, not the summary.

Does Tyler think AI will send people who love standup comedy a summary of the jokes made in a given standup, as a substitute for going to comedy shows?

🤖 Here’s your summary of this standup:

  • 3 jokes on women and stereotypes

  • 4 jokes on how clumsy he is

  • 2 playful racist jokes

  • 2 hecklers were addressed

  • Applause was 3/5 compared to other performers

We hope you’ve enjoyed this hilarious AI summary from ComixAI.

That doesn’t work for comedy, and it won’t work for young kids consuming viral/toxic content. The only way that can work is if they had a draconic blocker of all social media, and this was liked a service that summarized the content.

But it wouldn’t be something the kids would install themselves, nor happily use.

All that being said though, he is Tyler Cowen, so my chances of misunderstanding him and/or being wrong here are higher than usual.

Deepfaked Content Summaries (Click for thread)

AI and Music (Click for thread)


🔥⚒️Luke Stephens (hakluke) put out an amazing blog on his evolving approach to bug bounty automation. He talks about going from Bash, to Python, to Golang, and then arriving at Cloud-native. Great piece. MORE

🔥Thomas Roccia wrote an amazing piece on applying LLMs to Threat Intelligence. Includes a full notebook for running agents using Langchain that perform various TI tasks. MORE | THE CODE NOTEBOOK

🛠️SWE-agent - Autonomously fixes bugs in GitHub repos, showing a 12% success rate by leveraging language models like GPT-4. By Princeton Natural Language Processing | MORE

💻 Burr is a Python framework that simplifies building GenAI apps by managing state with easy-to-use building blocks. | by DAGWorks-Inc  | MORE

📖GIAM, a free open-source textbook, makes the art of mathematics accessible, covering foundational topics and proof techniques, complete with amusing chapter quotes. MORE

⚙️ Gram turns threat modeling into a self-hosted web app, making it easier for teams to collaborate on security. | by Klarna-Incubator | MORE

🔧 ChatGPT now lets you tweak images, like removing objects or changing parts, just by describing what you want in the chat. | MORE

🔧 Claude API's new "tools" feature lets it tap into the internet, opening up a world of data it couldn't touch before. | MORE

Kids are learning math from deepfake Taylor Swift and Drake on TikTok, and it's surprisingly effective. MORE

💡 I’m honestly super excited for deepfakes for these types of use-cases. Take the bad and find the good. Like, if kids absolutely love Taylor Swift, or Star Wars, let’s use that for education, tutoring, therapy, etc.

We need to be careful with it, of course, and watch out for IP issues, but there’s a lot of good that can come from this.


Check out one of suno.ai’s top AI-generated songs.

You might not like it, but if you don’t, how different is it from other stuff you don’t? Is it that much worse? And if you do like it, how much can you tell it’s not human-generated?


Don’t explain your philosophy. Embody it.


Thank you for reading.

UL is a personal and strange combination of security, AI, and lots of human content. 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 it with them. 🫶 

Happy to be sharing the planet with you,