2024-4-23 02:25:11 Author: danielmiessler.com(查看原文) 阅读量:13 收藏
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,
So Llama3 came out last week, and it’s really impressive. I’ve been playing with this new AI UI called Bolt.ai, which is quite nice. It’s basically a full application with a lot of the UX behavior of ChatGPT, but with the ability to use lots of different models.
There are many web versions of this type of thing, but this is much easier to install on a Mac.
Bolt.ai
Anyway, Llama3 has been pretty impressive at the 70B level. I haven’t done full testing yet, but I’ve had it generate at least a few responses that felt GPT-4-ish, and many that felt way worse. Remember that shaping open models with good system prompts is super important, and that going over the context window (8K for Llama3) makes it act crazy.
Also, Llama3 is significantly less restricted than previous models. In a lot of ways it behaves more like an uncensored model, especially if you tell it to act like one.
It’s insane to me that we’ll soon have GPT-4 level local models. Free. Local. And the resources required to run them will keep coming down. This is especially trippy when you realize that our standards for their performance will plateau for most tasks. Meaning, we’ll soon be able to do some massive percentage of everyday human tasks using local models that cost virtually nothing.
More stuff going on:
Continued prep for B-Sides/RSA shenanigans.
Remember to come say hi or should from across the room if you see me around BSides or RSA. Hugs, waves, finger guns, or fist-bumps all accepted, according to your preference. If I seem distracted, not very social, shy, introverted, awkward, etc., it’s because I am those things at that moment. Apologies. We can re-sync after.
My last few talks have gone extremely well. And one of them I didn’t even present that well due to some technical issues with the venue. There is just tremendous power in speaking to share an idea rather than trying to “execute a presentation” that people hopefully don’t think sucks. Night and day difference.
Updated the intro to the newsletter, focusing on Human 2.0 to Human 3.0. Let me know what you think by replying!
Another experiment this week: I sprinkled DISCOVERY into each of the SECURITY, TECH, HUMAN sections rather than being dedicated. Let me know what you think of that by replying as well! I like it because it’s clean, but don’t like it because it mixes news with links. Let me know your throughs.
🔥Oh, and you HAVE to go listen to this conversation between Tyler Cowen and Peter Thiel. I’m not a Peter Thiel fan because reasons but this conversation has caused me to re-think my assessment of his intelligence and understanding of the world. This conversation went from The Bible, to Shakespeare, to Star Wars, to the Antichrist. Seriously impressive. And if you’re wondering how I of all people could recommend Peter Thiel, see the Ideas section below. MORE
Ok, let’s get to it…
MY WORK
Wrote a new essay on how the old paradigm of planning a career no longer works. READ IT
SECURITY
The US House just passed a bill making it illegal for the government to buy your data without a warrant, calling it "The Fourth Amendment is Not For Sale." MORE
💡This is in response to people finding out that government agencies were just outright buying US citizens’ data from data brokers. I love this move.
Sandworm, a notorious Russian hacking group, has been linked to a cyberattack on a Texas water facility. MORE
The House just passed a bill that could ban TikTok in the U.S. if it's not sold of to a US company. MORE
MITRE was compromised by state-affiliated attackers using two Ivanti VPN zero-days. China-based attackers are suspected. MORE
A flaw in PuTTY versions 0.68-0.80 lets attackers with 60 cryptographic signatures from a user figure out their private keys offline. MORE
Sponsor
🔍Learn How to Demonstrate Secure AI Practices with ISO 42001
How are you proving your AI practices are secure? ISO 42001 was recently introduced to help companies demonstrate their security practices around AI, in a verifiable way.
Join Vanta and A-LIGN for a live session on May 14 to dig into ISO 42001— what it is, what types of organizations need it, and how it works.
Discover the components of the framework
Gain insights into which organizations can benefit most
Learn practical strategies and best practices for successfully integrating ISO 42001 into your organization
FBI Director Christopher Wray highlights an urgent shift in Chinese hacking strategies, saying they’re aiming to gain the ability to disrupt U.S. critical infrastructure by 2027 as part of prep for going into Taiwan. MORE
Moxie Marlinspike says he’s no longer affiliated with Signal.
💡I’ve never loved Signal so I’m going to be asking more people to switch back to Messages. Moxie was the only reason I saw it as equal or superior, and with him gone I see no reason to stay. MORE
Sponsor
VIRTUAL OPEN SOURCE POWERED SECURITY CONFERENCE
Join us for Hardly Strictly Security: The Ultimate Open Source Cybersecurity Conference. This Thursday, April 25th! This free, virtual conference is for security engineers, red teamers, bug bounty hunters, and security leaders. Hear from speakers from Vercel, Hashicorp, Datadog, Fastly, and others who have leveraged open source tools to make themselves - and all of us - more secure.
Sacramento International Airport had to stop flights due to a deliberately cut AT&T internet cable that provided internet to the airport. MORE
🔧 Tailscale SSH, now generally available, simplifies SSH by managing authentication and authorization. | by Tailscale | MORE
TECHNOLOGY
DeepMind's boss says Google's set to outspend everyone in AI, hinting at dropping over $100 billion into the tech. MORE
💡Outspending isn’t the same as outproducing or outshipping. The company has lost the ability to ship good products because they’re not guided by vision and customer needs anymore. They’re guided by an ancient GMail culture of engineers making stuff and throwing it at the wall to see if someone likes it.
I think they need a fresh start with new senior leadership.
Stanford's released a quality report on the state of AI models. Here’s a Fabric create_micro_summary: MORE
ONE SENTENCE SUMMARY:
The 2024 AI Index Report by Stanford University highlights AI's growing societal impact, technical advancements, and investment trends.
MAIN POINTS:
AI surpasses humans in specific tasks but not in complex reasoning and planning.
U.S. leads in AI model development with industry-dominating frontier research.
Investment in generative AI surged, reaching $25.2 billion in 2023.
TAKEAWAYS:
Training costs for top AI models are reaching unprecedented levels.
Lack of standardization in responsible AI evaluations complicates risk assessment.
AI's role in accelerating scientific progress and productivity is expanding
An interesting argument about how search engines, especially Google with its 90% market share, can sway election outcomes a lot more than we talk about. MORE
💡Another example of the power pendulum swinging back to companies.
Google fired 28 employees for protesting a $1.2 billion contract with Israel, citing policy violations and workplace disruption. MORE
Google merged its Android and hardware teams to innovate faster. MORE
Netflix runs FreeBSD CURRENT for its edge network due to a unique blend of stability and features. MORE
Reddit's showing up a lot more in Google results. MORE
Apple's AirPlay is starting to show up in hotel rooms. MORE
The TinySA is a budget-friendly spectrum analyzer. MORE
Programming is mostly thinking. MORE
A broad introduction to AWS logs sources and relevant events for detection engineering. | MORE
HUMANS
Generation Z is outperforming previous generations at their age. MORE
This article says societal decline mirrors the "Death Spiral" seen in ants, where companies and societies fall into self-destructive patterns, often ignoring early warning signs until it's too late to reverse the damage. MORE
Why Everything is Becoming a Game. MORE
A study found that jobs that require you to think a lot are protective against Alzeimer’s. MORE
Bayer is doing an experiment where they remove most of middle management and let 100,000 employees self-organize. They’re hoping it’ll save $2.15 billion. MORE
💡This will be another effect of AI. And I don’t mean AI tech, but AI’s influence on how to think about a business. AI implementations for businesses will look at everything in a business, from the products they’re making, the people they have, and the organizational structure, and recommend ways to massively improve efficiency by removing waste.
And that will often mean getting down to vision people and executors, with very little friction in between.
The term "brainwashing" morphed into a blanket term for any unconventional behavior in the US, sparking wild government experiments like MK-Ultra. MORE
IDEAS & ANALYSIS
Harvesting Ideas from Questionable People
This episode of the newsletter talks about Peter Thiel, who is basically one of the 7 anti-Christs in a lot of liberal circles. I dismissed him years ago because he supported Trump.
I feel like I’ve grown quite a bit in the last few years though. And I am conscious of making sure I just haven’t become more right-wing. I actually feel more grounded as a progressive than ever. Not a modern liberal, or leftist, but a progressive.
I guess my evolution is similar to Jonathan Haidt’s. He was super liberal before writing The Righteous Mind, which I highly recommend. It is the book that most influenced me to become a centrist. Not a move to the right, not a move away from the left, but something new.
The way I would describe it, which is not in that book, is to first define what you believe to be true, and the world you think we should live in. Don’t think about politics. Don’t think about parties. Those are all silly and ephemeral. Instead, imagine the actual society you would like to live in.
For me it’s something like (VERY raw/crude):
An understanding that evolutionary biology is the foundation of most tendencies and natural patterns for human and other animal societies
An understanding that we as humans can build on top of those tendencies to make something better
The lack of belief in libertarian (absolute) free will, such that criminals aren’t considered garbage, and billionaires aren’t considered gods
Free speech and free press, up to the point of actively/directly inciting violence against someone
It being both illegal and socially reprehensible to deny someone privileges because of their race or gender identity
Human first, tech second
Humanities first, sciences second
People’s reputations are harmed when they say things that are untrue
Simultaneous embrace of progressive and conservative ideas, accepting that for each given situation one might be better than the other to accomplish the goals of a given individual, family, or society.
A belief that most people are capable of being good and useful, if they’re properly supported when growing up
A belief that it is everyone’s responsibility to try to help everyone get that proper support growing up. Not technically, but as a society
Taxation is unpleasant but necessary, but we can’t let out of control government become so useless that it turns the rich against the idea
The rich (see lucky) see the raising of the poorest (least lucky) as not only good, but good for them as well
The primary goal of a member of society is to be useful
The successful (especially the self-made hustlers) are celebrated because hustle and usefulness are celebrated
Society is built on a blend of conservative ideas that respect our animal natures and progressive ideas that lift us beyond them, with the unifying factor being the lifting of all humans to lower amounts of suffering, and higher amounts of meaning and fulfillment
So let’s take those (but a better version, obviously, since I didn’t even use AI to write those out), and let’s say that’s our society.
Well, now I don’t care about liberal or conservative. Or right or left. Or any of those labels. They’re stuck in the current time, in the current Overton Window.
What we do instead is see parties—and the people within them—as idea sources. Because now I can discard good or bad ideas based on how they propel or distract from the world we’re trying to build.
And that brings me to Peter Thiel.
I discarded him because he supported Trump. Fair enough. Maybe he was dumb at the time. Maybe I was. That’s my own value judgement. But the point is that he could have changed (and I think I heard him say that actually).
But the point is that if I hear Peter Thiel say something smart, I’m going to listen. And if I hear him say something dumb, I’m going to stop listening.
Same for Joe Rogan. Or Andrew Huberman. Or even Sam Harris.
In an extreme form of this, if Ghengis Khan has the best bagel recipe on planet earth, I might use it. And if Peter Thiel wants to get Trump elected again, which I think would be horrible for the planet, but he also has something to teach me about political philosophy, I’m going to listen.
I. Will. Harvest. Good. Ideas.
My goal is to have the best models possible for how the world works. And if Peter Thiel or Ghengis Khan has better models than me for bagels, or supply side economics, then I will adopt them.
I can do this because I already know the society I want to help build. I know what goodness is. I know what evil is.
And because I have that footing, a bagel recipe isn’t going to somehow convince me to want a shittier society.
So, my recommendation…
RECOMMENDATION OF THE WEEK
Establish your ground truth in terms of morality and the society you want to live in. Lock that in without labeling it left, right, or whatever
Widely explore ideas from anyone and everyone
Do not discard people as a source of ideas just because you disagree with them on something, even if it’s major. That’s only hurting you, and the good you could do in the world as a result of being upgraded
Feel free to label people as overall bad, or stupid, but realize it doesn’t mean they’re wrong about everything. Example, I know know after seeing Tucker Carlson on Joe Rogan that Carlson is an actual idiot. Like, not a little bit. So I’ve closed my aperture to him largely, but not all the way. Again, if he has a great coffee recipe I’ll listen.
Regularly revisit your #1 and refactor everything
Regularly do #2
In short, don’t limit yourself by closing your ears to everyone who’s stupid about something. Most of us are.
And on that note, go listen to the conversation between Tyler Cowen and Peter Thiel. It was extraordinary, and it resulted in me buying a LOT of books. MORE
APHORISM OF THE WEEK
❝
The best way to have a good idea is to have a lot of ideas.
Linus Pauling
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,
如有侵权请联系:admin#unsafe.sh