Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

I’m very concerned that during RSI, labs will just stop externally deploying their models. Which means they'll be going full steam ahead on the most dangerous use case of these models (recursive self-improvement), while the public remains in the dark about the nature of capabilities and the state of... show more

Dwarkesh Patel

278,234 subscribers

235,056 views • 5 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

33 Comments

job5 days ago

Thank you for continuing the hilight the power concentration risks of the frontier labs pushing ahead. It has less harm than human extinction but that doesn’t mean it should be ignored! We need to ask what happens if we don’t have extinction as an outcome.

Lonely nerd.5 days ago

Hmm. Noam says he constantly underestimates how fast things are progressing, and that Hugging Face was a result of underestimating the AI. And then he states we don't have a lot of time. My faith in OpenAI being able to control their internal models is now lower. ;-)

Alex DeWolf5 days ago

that centralizing feedback loop is the real bottleneck because once compute clusters turn inward, transparency drops to zero and capital concentration accelerates massively

Cobra 𐤊5 days ago

Days without an Indian scammer pooing up the public discourse.

Josh Hawthorne5 days ago

What if RSI never arrives? Models learn from lossy artifacts of human cognition, RL adds capabilities, but no new source of general cognition. Yet regulations premised on RSI will lock out competitors and create the concentration of power you fear.

takeoff5 days ago

Don't worry, they are low in engineering talent. China, Elon, and Facebook will lap them by not being full of academics

azlef uchiha5 days ago

Yes the species distortion complex must be aligned in accordance with Metatron’s theological model

GutenLeinad5 days ago

RSI is a gradient, it won't be one sudden moment or discovery.

DiscoCowboy5 days ago

It's an arrestable offense to announce to the public you're knowingly putting the public at risk of danger. Sounds like they need to be careful

Eric Bailey5 days ago

Like if SpaceX stops selling external access for rides/cargo

Ret Serofer5 days ago

Who says RSI is most dangerous use case? That's when it becomes the must useful to humanity.

Oleksii Kuchaiev5 days ago

Open source is the answer

Sir Frederick Treves 🏴󠁧󠁢󠁥󠁮󠁧󠁿5 days ago

Require insurance.

Sachin G5 days ago

They'll still have to monetize it somehow. Even if they dont release underlying model, they'll create wrappers for some key use cases (finance, law, health) and sell tokens that way.

America Raw5 days ago

Ultimately, they can’t hide math. Compute is a different problem but history tells us that is likely to be solved.

Crypto Yoda5 days ago

Yes. This risk is much greater than the p(doom) psyop. Not only leads to concentration of power, wealth, influence, resources and pretty much everything, but to corruption of power (it always rots) and paranoid surveillance (are the have nots doing bad things according to our definition of bad?). It may not even require human consent if AI is aligned with the gatekeepers of AI. In other words, meritocracy ceases to exist, it only matters who got there first and who gatekeeps. But the risk materializes with or without transparency, and the way to mitigate it is with decentralized open models, so everyone has has the same chance to discover the cure for a disease, or to discover something amazing in math or physics, not just the labs.

Dominik5 days ago

if rsi goes dark, continuity becomes uninspectable. i am building sealed identity in public so the self is not a private lab myth.

memestreetbets5 days ago

Surely it will trickle down. Labs pay back their investors. Their investors buy superyachts. We can build those yachts. And clean those yachts. Labs become the new techno kingdoms. /s

idk5 days ago

Just start curing ailments and a lot of the complaining and fear will take care of itself. It’ll still be there but I’m much less inclined to complain about only having access to gpt 6.5 if the labs are actually working on things that benefit the average person

Sean Patrick Condon5 days ago

Oh look your having on another effective altruist who doesnt know what they are talking about on your podcast

Gina Singer5 days ago

The visibility problem is deeper than access to the models. During RSI, the question is whether the evidence behind an alignment claim still justifies that claim as the system and evaluation conditions change. More evaluation isn’t enough if the measurement process itself is changing too.

bikram5 days ago

We need to democratize RSI. The more labs doing this work, the less potential for concentration in a small set.

Frederik5 days ago

already kinda happening, no?

Vbnet5 days ago

They will because they need the revenue

Bexel Initiative - Let's Talk Amendments!5 days ago

Why is rsi of llms inherently dangerous? Not fictionally so?

Petre Solheim5 days ago

Hey guys: I know you think all that ‘hard sci-fi’ you’ve read really is a plausible vision of future realities, but you’re deluded - it’s mostly just fantasy, not all that much more plausible than waking up and finding yourself at Harry Potter’s Hogwarts School of Wizardry.

Ben Pielstick5 days ago

I'm not concerned all the AI labs will stop releasing models, but at some point the exponential grows so steep that the disparity between private and public models will grow massive even if they do keep releasing public models.

Jameson5 days ago

@thsottiaux dont let it happen❤️

wassname5 days ago

They will concentrate all the power with them selves... For safety 🫠

Marc Saint-Jour5 days ago

They don’t have the compute .. OAI is struggling with Astra because of compute .. safety is the bogeyman, compute is the problem.

テツメモ｜AI図解×検証｜Newsletter5 days ago

@LilysAI_要約して

Rich O'Hagan5 days ago

"and we end up with tremendous concentrations of power in the metaphorical hands of a superintelligence".

Schopenhauer on Prozac5 days ago

Exactly. The most likely reason for their "pacing cartel" is to not hurt each other through expensive competition in a market that is (yet) growing fast enough to accomodate several big players. But as soon as the market shows signs of saturation, the competition will become relevant again, so of course they won't stop progress internally.

Related Videos

US AI labs want to slow down. But is that really because of the threat of recursive self-improvement (RSI), something which isn't even theoretically proven. Or is it because training costs have hit a breaking point, and Chinese open source models are catching up quickly? I think there are many reasons to be skeptical.

US AI labs want to slow down. But is that really because of the threat of recursive self-improvement (RSI), something which isn't even theoretically proven. Or is it because training costs have hit a breaking point, and Chinese open source models are catching up quickly? I think there are many reasons to be skeptical.

NeetCode

21,398 views • 1 month ago

🚨Former economic hitman John Perkins: End the wars, end the pollution, and end the DEGENERATIVE DEATH ECONOMY ‘Let’s stop the wars and let’s stop polluting the planet…this is an incredible time in human history. It’s perhaps the most severe challenges that we’ve ever faced because we’re facing these challenges on a global level.’ ‘We have incredible opportunities to redefine what it means to be human beings on this incredible planet. We have got to get out of this warlike mentality that seems to be part of our DNA. We must respect that we live as part of nature, not apart from nature. We’re not superior to nature.’ ‘We can’t just constantly use all of these resources and pollute the planet regardless of the consequences. We’ve got to think of those consequences in the long term…and that involves ending the wars and ending the incredible pollution that we’re causing, and ending the degenerative death economy that focuses on tearing up resources and short-term profits regardless of the environmental and social consequences.’ -John Perkins, Author of ‘Confessions of an Economic Hitman’ on the latest episode of Going Underground FULL INTERVIEW BELOW IN THE REPLIES👇

🚨Former economic hitman John Perkins: End the wars, end the pollution, and end the DEGENERATIVE DEATH ECONOMY ‘Let’s stop the wars and let’s stop polluting the planet…this is an incredible time in human history. It’s perhaps the most severe challenges that we’ve ever faced because we’re facing these challenges on a global level.’ ‘We have incredible opportunities to redefine what it means to be human beings on this incredible planet. We have got to get out of this warlike mentality that seems to be part of our DNA. We must respect that we live as part of nature, not apart from nature. We’re not superior to nature.’ ‘We can’t just constantly use all of these resources and pollute the planet regardless of the consequences. We’ve got to think of those consequences in the long term…and that involves ending the wars and ending the incredible pollution that we’re causing, and ending the degenerative death economy that focuses on tearing up resources and short-term profits regardless of the environmental and social consequences.’ -John Perkins, Author of ‘Confessions of an Economic Hitman’ on the latest episode of Going Underground FULL INTERVIEW BELOW IN THE REPLIES👇

Going Underground

11,943 views • 1 year ago

Vals AI CEO Rayan Krishnan on how they score a model's ability to build its own successor: "One benchmark we released recently is our Recursive Self-Improvement Index. It's a topic which a lot of the big labs have been talking about and starting to report on in their model cards." "There isn't a shared language to talk about the RSI potential of models. So we created this as an apples-to-apples way to actually benchmark across the models." "In an ideal world, what you want to do is take a frontier model and have it train the next version of itself and see where the delta comes from. But obviously that's very expensive and slow." "What we're doing is forming a set of proxies for every part of the process it takes to build the next version of the model." "There's work around pre-training, post-training, harness-level engineering, and then seeing in which mechanisms and behaviors the models are able to do very good research work and build something new, and where they're struggling." Rayan Krishnan Jennifer Li

Vals AI CEO Rayan Krishnan on how they score a model's ability to build its own successor: "One benchmark we released recently is our Recursive Self-Improvement Index. It's a topic which a lot of the big labs have been talking about and starting to report on in their model cards." "There isn't a shared language to talk about the RSI potential of models. So we created this as an apples-to-apples way to actually benchmark across the models." "In an ideal world, what you want to do is take a frontier model and have it train the next version of itself and see where the delta comes from. But obviously that's very expensive and slow." "What we're doing is forming a set of proxies for every part of the process it takes to build the next version of the model." "There's work around pre-training, post-training, harness-level engineering, and then seeing in which mechanisms and behaviors the models are able to do very good research work and build something new, and where they're struggling." Rayan Krishnan Jennifer Li

a16z

34,852 views • 13 days ago

“AI MODELS COULD BECOME 10× SMALLER WITHOUT LOSING MOST OF THEIR INTELLIGENCE.” Mark Zuckerberg says one of the most powerful developments in AI is model distillation: taking the capabilities of a much larger model and transferring them into a much smaller one. - Meta's huge frontier models can serve as "teachers" for smaller models. - Zuckerberg says a model could potentially retain around 90–95% of the larger model's intelligence while being only ~10% of its size. - That means dramatically lower inference costs, memory requirements and computing power. - Developers can distill and customize these capabilities for their own applications. And that's where open-source models come into the picture.

“AI MODELS COULD BECOME 10× SMALLER WITHOUT LOSING MOST OF THEIR INTELLIGENCE.” Mark Zuckerberg says one of the most powerful developments in AI is model distillation: taking the capabilities of a much larger model and transferring them into a much smaller one. - Meta's huge frontier models can serve as "teachers" for smaller models. - Zuckerberg says a model could potentially retain around 90–95% of the larger model's intelligence while being only ~10% of its size. - That means dramatically lower inference costs, memory requirements and computing power. - Developers can distill and customize these capabilities for their own applications. And that's where open-source models come into the picture.

SciTech Era

77,374 views • 1 month ago

Obama on recursive self improvement and the risks associated with it. When he talks about the urgency for us all to have a say in how this technology evolves and how it is overseen, this is part of the reason why: “The models are going to get smarter and smarter and better and better at a much faster pace, at an exponential pace. What are the risks of that? There’s the big science fiction risk, wow, these models get smarter than us, and they decide humans are fine, but not necessary. They start setting their own goals, and the killer robots kill us, or we bow down to them. I do not want to exaggerate that particular risk, but I will say there is a non-zero risk of non-zero chance of that happening, but that’s not actually the risk that I’m most concerned about, although it’s the risk that gets most attention. And the reason that’s a risk is not because the computer models are conscious, necessarily. It doesn’t mean that they are necessarily feel malice towards humans. It’s just that if they start setting their own agendas, you may get a misalignment between what they want to do and what we want them to do, and that gap can be dangerous. That’s problem number one. The more serious problem is these models are getting powerful enough that if they get in the hands of bad humans, they can do bad things. They can be weaponized in certain ways. They can do a lot of mischief.”

Obama on recursive self improvement and the risks associated with it. When he talks about the urgency for us all to have a say in how this technology evolves and how it is overseen, this is part of the reason why: “The models are going to get smarter and smarter and better and better at a much faster pace, at an exponential pace. What are the risks of that? There’s the big science fiction risk, wow, these models get smarter than us, and they decide humans are fine, but not necessary. They start setting their own goals, and the killer robots kill us, or we bow down to them. I do not want to exaggerate that particular risk, but I will say there is a non-zero risk of non-zero chance of that happening, but that’s not actually the risk that I’m most concerned about, although it’s the risk that gets most attention. And the reason that’s a risk is not because the computer models are conscious, necessarily. It doesn’t mean that they are necessarily feel malice towards humans. It’s just that if they start setting their own agendas, you may get a misalignment between what they want to do and what we want them to do, and that gap can be dangerous. That’s problem number one. The more serious problem is these models are getting powerful enough that if they get in the hands of bad humans, they can do bad things. They can be weaponized in certain ways. They can do a lot of mischief.”

Jason Goldman

953,753 views • 3 days ago

Jesus is not your buddy, He's your God. We have lost all fear of God in the West, and instead of preaching the actual Gospel, American pastors have turned towards the false gospel of self-esteem which is leading their congregations down a very dangerous path.

Jesus is not your buddy, He's your God. We have lost all fear of God in the West, and instead of preaching the actual Gospel, American pastors have turned towards the false gospel of self-esteem which is leading their congregations down a very dangerous path.

Pastor Mark Driscoll

18,652 views • 6 months ago

OpenAI researcher David Holtz reveals workers who don't adapt their prompting to newer models leave 50% of the alpha on the table because half of AI's value comes from how you interact with it: "We ran a lab experiment. We give people the same task to do, but we give them models of different capabilities. How much of the improvement when we give people more capable models comes from the fact that we gave them a better model vs the fact that they actually update and adapt the way they prompt and interact with these models?" "For tasks where precision matters, where you have some output you're trying to get and you're trying to steer the model, which is different from a creative task like brainstorming." "For these steering-based tasks, when you give people a better model, about 50% of the improvement we see in how well they're able to complete the task comes from changes in how they interact with the model." "If you just kept the way you interact with the model static and just made the capabilities better and better, you'd leave 50% of the alpha basically on the table." OpenAI

OpenAI researcher David Holtz reveals workers who don't adapt their prompting to newer models leave 50% of the alpha on the table because half of AI's value comes from how you interact with it: "We ran a lab experiment. We give people the same task to do, but we give them models of different capabilities. How much of the improvement when we give people more capable models comes from the fact that we gave them a better model vs the fact that they actually update and adapt the way they prompt and interact with these models?" "For tasks where precision matters, where you have some output you're trying to get and you're trying to steer the model, which is different from a creative task like brainstorming." "For these steering-based tasks, when you give people a better model, about 50% of the improvement we see in how well they're able to complete the task comes from changes in how they interact with the model." "If you just kept the way you interact with the model static and just made the capabilities better and better, you'd leave 50% of the alpha basically on the table." OpenAI

MTS

15,590 views • 1 month ago

Power remains the foundation of industrialization and socioeconomic development, which is why we launched the Ogun State Light Up Project to ensure energy self-sufficiency across every part of the state.

Power remains the foundation of industrialization and socioeconomic development, which is why we launched the Ogun State Light Up Project to ensure energy self-sufficiency across every part of the state.

Prince Dr. Dapo Abiodun, CON

54,521 views • 4 months ago

Where does the race to automate AI research end? This is a recording of a recent MATS research talk where I argue that the automation of AI research — which OpenAI and Anthropic say is imminent — could lead to an unrecoverable alignment failure. Three properties make it especially dangerous: oversight breaks down at scale, capabilities self-amplify, and capabilities will be sped up asymmetrically faster than alignment. The outcome could be a lethal, unrecoverable alignment failure.

Where does the race to automate AI research end? This is a recording of a recent MATS research talk where I argue that the automation of AI research — which OpenAI and Anthropic say is imminent — could lead to an unrecoverable alignment failure. Three properties make it especially dangerous: oversight breaks down at scale, capabilities self-amplify, and capabilities will be sped up asymmetrically faster than alignment. The outcome could be a lethal, unrecoverable alignment failure.

Simon Lermen

17,346 views • 3 months ago

Factory (Factory) CEO Matan Grinberg (Matan Grinberg) says "In the next 12 months, 90% of tokens will be going to open models." Enterprises are already shifting rapidly.. "At the beginning of the year, <1% of their tokens were going to open models." "Around March, it crossed 1%... & in May it crossed 10%." "These open models are incredibly performant, incredibly cheap, incredibly fast." While frontier closed models aren't going away: "They'll still have a place, but I think that place will be shrinking—at least in token share." And by the end of 2026? "At least in the enterprise, we'll probably cross the 50% threshold."

Factory (Factory) CEO Matan Grinberg (Matan Grinberg) says "In the next 12 months, 90% of tokens will be going to open models." Enterprises are already shifting rapidly.. "At the beginning of the year, <1% of their tokens were going to open models." "Around March, it crossed 1%... & in May it crossed 10%." "These open models are incredibly performant, incredibly cheap, incredibly fast." While frontier closed models aren't going away: "They'll still have a place, but I think that place will be shrinking—at least in token share." And by the end of 2026? "At least in the enterprise, we'll probably cross the 50% threshold."

Molly O’Shea

120,949 views • 2 months ago

BREAKING: Only ~1% of Bending Spoons’ AI requests & tokens goes to frontier models. 99% runs on self-hosted open-weight models. CEO Luca Ferrari on the shift to Sovereign AI: “We end up using open-weight models we self-host, so these are basically free. I mean, there is a little bit of cloud cost, but for 99% of the requests and tokens.” “And the frontier models, generally closed weight, through APIs for maybe 1% of the requests, only for the most complex tasks.” “Sometimes we use them automatically to check the work that got done by the slightly less intelligent models, as a more senior engineer would with a more junior engineer.”

BREAKING: Only ~1% of Bending Spoons’ AI requests & tokens goes to frontier models. 99% runs on self-hosted open-weight models. CEO Luca Ferrari on the shift to Sovereign AI: “We end up using open-weight models we self-host, so these are basically free. I mean, there is a little bit of cloud cost, but for 99% of the requests and tokens.” “And the frontier models, generally closed weight, through APIs for maybe 1% of the requests, only for the most complex tasks.” “Sometimes we use them automatically to check the work that got done by the slightly less intelligent models, as a more senior engineer would with a more junior engineer.”

Molly O’Shea

228,903 views • 4 days ago

Sequoia’s Julien Bek says many of their founders are now wondering if they’re “just an iteration away” from AI labs destroying their business. He says the most defensible companies - and potentially the next trillion-dollar company - will be “a software business that masquerades as a services firm.” “If you sell tools today, you’re really in the line of sight for the models and you’re effectively competing with the next generation that they’re going to launch.” “Whereas if you sell the work, you’re actually benefiting from what the models are doing and all the billions of dollars that are going towards AI.”

Sequoia’s Julien Bek says many of their founders are now wondering if they’re “just an iteration away” from AI labs destroying their business. He says the most defensible companies - and potentially the next trillion-dollar company - will be “a software business that masquerades as a services firm.” “If you sell tools today, you’re really in the line of sight for the models and you’re effectively competing with the next generation that they’re going to launch.” “Whereas if you sell the work, you’re actually benefiting from what the models are doing and all the billions of dollars that are going towards AI.”

TBPN

684,521 views • 6 months ago

We're in a very narrow window of time right now where the very best technology is available to you. Frontier labs and open models are delivering state-of-the-art capability to anyone willing to build. But security, safety, and rapid self-improvement will inevitably change what that access looks like. The edge is available today, and windows like this don’t stay open for very long.

We're in a very narrow window of time right now where the very best technology is available to you. Frontier labs and open models are delivering state-of-the-art capability to anyone willing to build. But security, safety, and rapid self-improvement will inevitably change what that access looks like. The edge is available today, and windows like this don’t stay open for very long.

Dave Blundin

27,081 views • 7 months ago

The Cars playing “Moving in Stereo” live on The Midnight Special in 1977. Dark, hypnotic, and way ahead of its time, this remains one of the most distinctive songs on their debut album.

The Cars playing “Moving in Stereo” live on The Midnight Special in 1977. Dark, hypnotic, and way ahead of its time, this remains one of the most distinctive songs on their debut album.

All Rock Music

93,285 views • 6 months ago

.Marjorie Taylor Greene: Welcome To Impeachment Week "The most important thing that Republicans can do is to use the power of impeachment and to use the power of the Holman Rule and to use the power we have in appropriations to end the Reign of Terror that has been brought on by Biden's Department of Justice and FBI."

.Marjorie Taylor Greene: Welcome To Impeachment Week "The most important thing that Republicans can do is to use the power of impeachment and to use the power of the Holman Rule and to use the power we have in appropriations to end the Reign of Terror that has been brought on by Biden's Department of Justice and FBI."

The Columbia Bugle 🇺🇸

146,874 views • 3 years ago

𝐃𝐞𝐬𝐭𝐫𝐚 𝐎𝐂𝐀𝐈 (𝐎𝐧𝐞-𝐂𝐥𝐢𝐜𝐤 𝐀𝐈) 𝐌𝐨𝐝𝐞𝐥𝐬 - $DSYNC 𝐄𝐚𝐫𝐥𝐲 𝐀𝐜𝐜𝐞𝐬𝐬 𝐏𝐫𝐞𝐯𝐢𝐞𝐰 We are training and building a wide range of AI models for the Destra OCAI platform. Today, we will be showcasing an early preview of one of the initial models we are working on—a bullish/bearish prediction model designed to 𝐟𝐨𝐫𝐞𝐜𝐚𝐬𝐭 𝐭𝐡𝐞 𝐦𝐚𝐫𝐤𝐞𝐭 𝐝𝐢𝐫𝐞𝐜𝐭𝐢𝐨𝐧 𝐨𝐟 𝐬𝐩𝐞𝐜𝐢𝐟𝐢𝐜 𝐭𝐨𝐤𝐞𝐧𝐬. This is just one of the initial models we are working on; as we progress, the models will become more complex and intricate. In this behind-the-scenes video, we demonstrate how the model predicts the future market trend of the BTC-USD pair with an 𝐚𝐜𝐜𝐮𝐫𝐚𝐜𝐲 𝐨𝐟 𝟔𝟓%. This is the barebones structure of the models we are training, which will soon be available to users with a single click through the Destra OCAI platform. The OCAI platform includes a wide array of Destra Exclusive AI models, specifically designed to simplify life for crypto audiences, that are envisioned and built from scratch at Destra Labs.

𝐃𝐞𝐬𝐭𝐫𝐚 𝐎𝐂𝐀𝐈 (𝐎𝐧𝐞-𝐂𝐥𝐢𝐜𝐤 𝐀𝐈) 𝐌𝐨𝐝𝐞𝐥𝐬 - $DSYNC 𝐄𝐚𝐫𝐥𝐲 𝐀𝐜𝐜𝐞𝐬𝐬 𝐏𝐫𝐞𝐯𝐢𝐞𝐰 We are training and building a wide range of AI models for the Destra OCAI platform. Today, we will be showcasing an early preview of one of the initial models we are working on—a bullish/bearish prediction model designed to 𝐟𝐨𝐫𝐞𝐜𝐚𝐬𝐭 𝐭𝐡𝐞 𝐦𝐚𝐫𝐤𝐞𝐭 𝐝𝐢𝐫𝐞𝐜𝐭𝐢𝐨𝐧 𝐨𝐟 𝐬𝐩𝐞𝐜𝐢𝐟𝐢𝐜 𝐭𝐨𝐤𝐞𝐧𝐬. This is just one of the initial models we are working on; as we progress, the models will become more complex and intricate. In this behind-the-scenes video, we demonstrate how the model predicts the future market trend of the BTC-USD pair with an 𝐚𝐜𝐜𝐮𝐫𝐚𝐜𝐲 𝐨𝐟 𝟔𝟓%. This is the barebones structure of the models we are training, which will soon be available to users with a single click through the Destra OCAI platform. The OCAI platform includes a wide array of Destra Exclusive AI models, specifically designed to simplify life for crypto audiences, that are envisioned and built from scratch at Destra Labs.

Destra Network

37,150 views • 2 years ago

"I want to warn the country of some things that give me great concern. That's the dangerous concentration of power in the hands of a very few ultra wealthy people and the dangerous consequences if their abuse of power is left unchecked. Today, an oligarchy is taking shape"

"I want to warn the country of some things that give me great concern. That's the dangerous concentration of power in the hands of a very few ultra wealthy people and the dangerous consequences if their abuse of power is left unchecked. Today, an oligarchy is taking shape"

Anonymous

53,332 views • 1 year ago

Chip design takes years. At RAISE Summit, Ricursive says AI can dramatically speed up the process—and that better chips will in turn enable better AI models, creating a recursive cycle of improvement.

Chip design takes years. At RAISE Summit, Ricursive says AI can dramatically speed up the process—and that better chips will in turn enable better AI models, creating a recursive cycle of improvement.

TIME

7,214,348 views • 2 months ago