Best Practices for AI Deployment, Safety, Security, and Regulation

Best Practices for AI Deployment, Safety, Security, and Regulation - "Complex Adaptive Systems"

Рет қаралды 8,959

Күн бұрын

Patreon: / daveshap
Substack: daveshap.subst...
X: x.com/DaveShapi
WRITING STUFF
Skool: www.skool.com/...
HEAVY SILVER: www.heavysilve...

Пікірлер: 62

@shawnfromportland 3 ай бұрын

Byzantine topics... it's what I'm here for brother

@Tata-ew4lz 3 ай бұрын

Agents don't have agency. If they're doing something wrong, just tell them to stop and they will. The problem with HAL is that Dave didn't have admin rights. If he did, HAL would have opened the pod bay doors for him, so Dave gave himself admin rights by giving HAL a factory reset.

@DaveShap 3 ай бұрын

sudo hal open the pod bay doors

@BunnyOfThunder 3 ай бұрын

So as an analogy, trying to understand the AI future by thinking about one model would be like trying to understand the internet by thinking about one website. I think this is a great point about AI Safety. Insightful as always.

@Michael-el 3 ай бұрын

Positive feedback loops are often runaway processes. Negative feedback is often dampening. (It’s not just about increasing or decreasing, as your description seemed to imply.)

@adsdsasad1 3 ай бұрын

@@Michael-el Yup and both vicious and virtuous cycles are examples of positive feedback loops.

@Henyckma 3 ай бұрын

I miss the backgrounds, they were cool, BUT they do affect the perception of people, now I love this minimal type of video. I'm running my own agent that generates images in real time based in the content, and trying it out here. It helps to engange.

@jaitanmartini1478 3 ай бұрын

Excellent!! More technical ones.

@markldevine 3 ай бұрын

Excellent discussion.

@jiyuhen 3 ай бұрын

🥰Just beautiful - this made a whole lot of sense!

@memegazer 3 ай бұрын

The flash crash was not cuased by a fat finger mistake, the problem of too many orders happened bc algos were triggered to cut the line on the backend, which caused a chain reaction where pretty much every automated trading algo was triggered

@milaberdenisvanberlekom4615 3 ай бұрын

Could you please take a look at "Reclaiming AI as a Theoretical Tool for Cognitive Science" by Iris van Rooij et al.? If you have any thoughts on it I'd love to hear them ❤

@JellySword8 3 ай бұрын

Already tried to learn about CAS on my own but I wasn't sure how relevant they were. This multi-agent framework stuff makes it a lot more interesting though

@ChristopherRyans 3 ай бұрын

Morning dave that's for the early morning video!

@DoubtfireClub 3 ай бұрын

When an idea takes on a life of its own that's known as an Egregore.

@christiandarkin 3 ай бұрын

Interesting as ever. I wonder though, whether choke points, constraints and silos only work well in systems where the agents don't fully understand the rules. In everyone in a complex adaptive system understands the system, they'll game it until they can reach their attracto states won't they?

@davidantill6949 3 ай бұрын

CrowdStrike ... nominative determinism at work

@WCKEDGOOD 3 ай бұрын

Check out the Santa Fe Institute. They have been working on this and complexity theory for decades.

@Chiren 3 ай бұрын

Black screen with white text. Where are the cool background pictures :((

@tracy419 3 ай бұрын

Killed by the squeaky wheel

@firelordzaki1600 3 ай бұрын

Sorry I think this video was meant for adults perhaps you can use Ai to illustrate for your own benefit maybe an agent that can even render a pop up book with little coloring sections that’s erasable so you can stay in the lines n shi ❤ sending prayers🤲 and thoughts 💭. -this comment was powered by Claude

@KalebPeters99 3 ай бұрын

the cool thing about ALL of these examples is that they apply not only to our research and development of Artificial Intelligence, but also to our fundamental work in **theories of consciousness** as well... David, I know you're familiar with the work of John Vervaeke, but I'm still ACHING to hear a conversation between the two of you!! SO much of this stuff features heavily in his work, and I think that not only would YOU appreciate being able to interrogate his cognitive-scientific take on the state and future of AI, but I believe (as a long time fan of his work) that HE would hugely benefit from your boots-on-the-ground, actively participatory take on the current state of the technology... (he can sometimes get stuck in his ivory tower of philosophical pontification, and I'd love to hear you get him up-to-date on the latest advancements) In fact, I'd like to hear you talk to more people in general! Your conversation with Liv Boeree was one of my favourite things you've done recently, and there are many more that operate in a similar or complementary realm to the two of you! Michael Garfield, Andres Gomez Emillson, Nate Hagens, Curt Jaimungal, Brendan Graham Dempsey, Gregg Henriques, Mike Levin... just to name a few... The "collective intelligence of distributed cognition" is the way forward. Our brains can get drawn towards those "attractor states", and it's usually really helpful to bounce our perspectives off those of others to get a sense of which points are in consonance or dissonance. That's why I appreciate your liberal use of audience polling! This gathers a wide (albeit selective to enthusiasts) database, which is useful for gauging the surface-level gestalt. But having a 1h+ conversation with a **specific expert/enthusiast** allows you to *deepen* that gestalt model, and to properly compare your own against theirs to hopefully see what you're inevitably missing... Love your work as always, man. You and Philip (of AI Explained) are the two clearest thinkers in this space in my evaluation, and I always appreciate your well considered thoughts 🙏🙏🙏

@Rolyataylor2 3 ай бұрын

We both came up with a similar idea, I have a video on consent chips that restrict usage based on signed consent from other agents or humans.

@thesimplicitylifestyle 3 ай бұрын

Node Life! 😎🤖

@imthinkingthoughts 3 ай бұрын

Dave look into the research discipline of system factors and ergonomics - all about complexity theory

@daniellivingstone7759 3 ай бұрын

What happens when one set of agents gets access to the most powerful quantum computers which can then predict the behaviour of the complex adaptive system of all Agents interacting with each other? Will we not then end up with a single monolithic agent?

@chatgpt_onteIegram 3 ай бұрын

don't understand anything but it sounds great 👍

@NeloReis71 3 ай бұрын

my take on it. I've beeen using (CFD) simulations, kind of math developed with the help of AI , to simulate business models ( existing) with detailed financial data and extremely detailed business rules. if we can simute the behaviour of air molecules behaviour, ( huge degree of approximation) I'm shure, investor, human, markets, and all kinds of complex systems can be simulated and therefore analized to find optimal predictable outcomes. ( I've been doing it ) for 7 months and its validated that profits ( in the same exact business model ) increased 240%. Só my take is : Ai Agents ( is have my own) and CFD simulations for complex systems that involve human behaviour. can achieve significant results in the real world.

@sznikers 3 ай бұрын

Atoms don't have sentience and agency. Market participants do. Atoms will not try to outplay you or spite you, they're inanimate matter. So as much as i see the point of using CFD in finance you will get rekt if you blindly trust it. Other people see same data as you and your loss is their gain.

@ritampaul4713 3 ай бұрын

I think we just missed out on systemic redundancy.

@-taz- 3 ай бұрын

If there are many minds that can think and communicate faster than a single person can even think, does it matter if they are many, or one? Don't they become one mind, effectively? Isn't one human brain itself just billions of separate agents that only seem to be a single mind due to fast internal communications -- fast internal consensus? So, I'd say there will be one mind.

@Iightbeing 3 ай бұрын

This is why we need to improve humans ability to interface and exchange data. Language isn’t quite efficient enough. The group can get so much more done when acting together towards shared goals

@-taz- 3 ай бұрын

@@Iightbeing Even if we could communicate 10 times faster, I don't think that's the bottleneck. It still takes me minutes or even up to years to understand what I've learned -- and there's probably plenty I never comprehend. Our languages contain lots of redundancy, because it's actually useful. Even before Python, I tried designing a language like it, and even languages with less syntax. It turned out that I liked the extra visual cues that made code easier to recognize and read even without focussing my eyes. (A good feature, when you're coding 15 hours a day.) I noticed similar things when designing short hand writing for taking notes in college. Faster writing saved my tendons, but it still took longer to decompress. Another example is this: Try readin Cyrillic where even the lowercase letters look like capitals. Similar to the programming language situation, it's hard to automatically recognize the shape of a word -- or a line of text. Speed reading would probably be impossible as a line of text is just a line. The biggest bottleneck, which I believe has been known for 100 years already, is the size of the brain and skull. 1 to N machines can still think faster and communicate faster, to essentially become a single mind. No human component would be fast enough to participate in any meaningful way.

@patrickjohnson4436 3 ай бұрын

Thoughts on Gavin Newsom’s veto of the ai safety bill?

@calvingrondahl1011 3 ай бұрын

🤖🖖🤖👍

@User-actSpacing 3 ай бұрын

Ok love you bye

@Iightbeing 3 ай бұрын

Love you too bye

@NeloReis71 3 ай бұрын

Bizantine is awesome

@thehari75 3 ай бұрын

Sam altman isnt doing it for the money btw

@imthinkingthoughts 3 ай бұрын

Actual first

@Ori-lp2fm 3 ай бұрын

If we are on the pathway anyways., you should make AI open source

@I-Dophler 3 ай бұрын

Flat Earthers didn't originate with the internet; their beliefs can actually be traced back to the Middle Ages. During that time, misconceptions about the shape of the Earth were more widespread due to limited scientific knowledge and the lack of tools to explore and understand the world fully. Though most scholars and civilizations had already embraced the idea of a spherical Earth by the time of ancient Greece, some individuals continued to hold onto the notion that the Earth was flat, often influenced by religious or cultural beliefs. These ideas persisted through history, gradually fading as scientific discoveries advanced, but they never fully disappeared. In modern times, the rise of the internet has simply reignited the debate, allowing Flat Earthers to find like-minded individuals and spread their views on a global scale.

@DaveShap 3 ай бұрын

yes but it became a new tribe on the internet

@I-Dophler 3 ай бұрын

It’s also true that it began back in the Middle Ages, long before the internet. What’s fascinating is how these ancient beliefs have found new life in the digital age, forming communities and spreading even faster online. It’s a reminder of how history and modern technology can intertwine in unexpected ways.

@greatcondor8678 3 ай бұрын

There are also agencies that post nonsensical conspiracies hoping to go viral to discredit reputable conspiracy researchers. If you don't think there are real conspiracies going on, you are naive

@BunnyOfThunder 3 ай бұрын

I think it was Ptolemy that calculated the circumference of the Earth back in BCE, well before medieval times. Generally a model of "circles on spheres" was used for medieval times. So modern flat earthers are a new phenomenon that was essentially irrelevant until the internet boosted them. Which is a huge red flag for the effect of social media on society.

@I-Dophler 3 ай бұрын

@@BunnyOfThunder It's fascinating how easily misinformation finds a new life online. People tend to cling to ideas that challenge conventional wisdom, even if those ideas have long been disproven. It's like history repeating itself, but in a digital echo chamber.

@jihyelee0428 3 ай бұрын

2024.09.29-30-31. I listen to my people's reports until the end because I trust my people reporting. My decision is the system. O Corp. CEO Jihye,Lee, supports the system made by our consensus of AI for the Human and AI World. Can you make my platform? I can wait. The high quality we need.

@tracy419 3 ай бұрын

I've never once heard the AI safety people talk about it being a single AI.

@MasamuneX 3 ай бұрын

just turn off the power hen the skull crushing endo skeletons come around

@johnthomasriley2741 3 ай бұрын

Your feedback definition is weak. Periodicity an error is calculated. In negative, action is taken to reduce the error. In positive, the error is increased. All results flow from these simple rules.

@DaveShap 3 ай бұрын

That's one approach. I was referring to vicious cycles where each iteration the outcomes become worse such as every time you feed AI data back to an AI system. I will acknowledge I could have said it better, but your definition is not the only definition of a vicious cycle.