I am convinced that only about 2% of users on social media platforms who actively create content are genuine humans, while another 18% are genuine users who primarily consume content. The remaining 80% are AI bots.
Every country should block access to social media platforms for six months leading up to major elections. AI-driven bot activity has turned these platforms into instruments of state-sponsored election interference, posing a serious threat to national security and democratic integrity.
Every country should block access to social media platforms for six months leading up to major elections. AI-driven bot activity has turned these platforms into instruments of state-sponsored election interference, posing a serious threat to national security and democratic integrity.
0
0
0
53
a few esoteric yet interesting use cases of ai i spent some time exploring this weekend.
1. sports and esports analytics
leveraging a combination of multimodal system 1 and system 2 models to review hundreds or thousands of hours of gameplay and understand how individual e.
e.g. league of legends. how does a mid laner鈥檚 behavior change when their jungler is nearby? do they become more aggressive? do these tendencies drift over time? what factors influence them?
or football (soccer). which direction does a player typically shoot during penalties? which direction does the goalkeeper tend to dive? are these behaviors dynamic or habitual? do they change under pressure?
imagine combining fast perception models with deeper reasoning models to identify behavioral patterns that even the players themselves might not be conscious of.
the second-order effects are perhaps even more interesting, particularly for sports betting.
if we can model individual player behavior with sufficient accuracy, what happens to the betting markets?
historically, sophisticated analytics and quantitative modeling have largely been concentrated among well-capitalized firms and specialized trading operations.
what happens when anyone can deploy intelligent agents to analyze thousands of matches, identify behavioral tendencies, and discover potential mispricings in betting odds?
of course, as these capabilities proliferate, bookmakers will incorporate similar intelligence, and whatever predictive advantage exists may quickly disappear.
an interesting arms race between those predicting human behavior and those pricing it.
2. simulators and digital twins
i鈥檝e also been playing around with the notion of using ai and reinforcement learning to construct intelligent simulations of complex real-world systems.
one idea is building a digital twin of capital markets, populated by thousands of agents interacting through simulated order books and incorporating real-time news, macroeconomic signals, and market activity.
another is creating an isolated simulated world for a company like apple, modeling its suppliers, competitors, consumer demand, pricing dynamics, and economic dependencies.
what happens if a supplier fails? tariffs increase? a competitor introduces a breakthrough technology? how do these changes propagate through the ecosystem?
the challenge, of course, is building simulations that actually reflect reality rather than simply produce convincing outcomes.
what fascinates me about both ideas is the possibility of moving beyond analyzing what happened to understanding why it happened, and experimenting with what might happen next.
a lot more to explore here than simply asking llms questions
1. sports and esports analytics
leveraging a combination of multimodal system 1 and system 2 models to review hundreds or thousands of hours of gameplay and understand how individual e.
e.g. league of legends. how does a mid laner鈥檚 behavior change when their jungler is nearby? do they become more aggressive? do these tendencies drift over time? what factors influence them?
or football (soccer). which direction does a player typically shoot during penalties? which direction does the goalkeeper tend to dive? are these behaviors dynamic or habitual? do they change under pressure?
imagine combining fast perception models with deeper reasoning models to identify behavioral patterns that even the players themselves might not be conscious of.
the second-order effects are perhaps even more interesting, particularly for sports betting.
if we can model individual player behavior with sufficient accuracy, what happens to the betting markets?
historically, sophisticated analytics and quantitative modeling have largely been concentrated among well-capitalized firms and specialized trading operations.
what happens when anyone can deploy intelligent agents to analyze thousands of matches, identify behavioral tendencies, and discover potential mispricings in betting odds?
of course, as these capabilities proliferate, bookmakers will incorporate similar intelligence, and whatever predictive advantage exists may quickly disappear.
an interesting arms race between those predicting human behavior and those pricing it.
2. simulators and digital twins
i鈥檝e also been playing around with the notion of using ai and reinforcement learning to construct intelligent simulations of complex real-world systems.
one idea is building a digital twin of capital markets, populated by thousands of agents interacting through simulated order books and incorporating real-time news, macroeconomic signals, and market activity.
another is creating an isolated simulated world for a company like apple, modeling its suppliers, competitors, consumer demand, pricing dynamics, and economic dependencies.
what happens if a supplier fails? tariffs increase? a competitor introduces a breakthrough technology? how do these changes propagate through the ecosystem?
the challenge, of course, is building simulations that actually reflect reality rather than simply produce convincing outcomes.
what fascinates me about both ideas is the possibility of moving beyond analyzing what happened to understanding why it happened, and experimenting with what might happen next.
a lot more to explore here than simply asking llms questions
0
0
1
120
this barista at @Starbucks got my order wrong 3 times...at this point i can't tell if its weaponized incompetence or just racism
0
0
0
26
within five years, owning an intelligence will feel as ordinary as owning a phone. you will own the model itself, the weights that give it shape.
it will arrive with knowledge and tools. then, through everyday use, it will begin to change. continual learning will weave your experiences and memories into its weights. the life you live will leave an imprint on the intelligence beside you.
two people might buy the same model and, years later, possess something entirely different. each will have grown around a different life.
many people will own several. one for work. one for their private world. one devoted to a project that takes a lifetime. we will choose which parts of ourselves each is allowed to know.
the model companies will resemble the phone and computer companies of today. they will sell the foundations. some purchases may arrive as nothing more than an api address, an intelligence you own and can reach through whatever interface you prefer.
the company will supply the beginning. living with it will shape what follows.
and slowly, another world will grow within this one: a world of minds, memories, and places made of computation.
in time, we will devote a hundred times more energy to sustaining that world than to the rest of the physical economy.
it will arrive with knowledge and tools. then, through everyday use, it will begin to change. continual learning will weave your experiences and memories into its weights. the life you live will leave an imprint on the intelligence beside you.
two people might buy the same model and, years later, possess something entirely different. each will have grown around a different life.
many people will own several. one for work. one for their private world. one devoted to a project that takes a lifetime. we will choose which parts of ourselves each is allowed to know.
the model companies will resemble the phone and computer companies of today. they will sell the foundations. some purchases may arrive as nothing more than an api address, an intelligence you own and can reach through whatever interface you prefer.
the company will supply the beginning. living with it will shape what follows.
and slowly, another world will grow within this one: a world of minds, memories, and places made of computation.
in time, we will devote a hundred times more energy to sustaining that world than to the rest of the physical economy.
0
0
1
32
I believe we will eventually create entire simulated worlds, perhaps universes, because aligning intelligence requires understanding how it develops within a civilization.
Values become meaningful when they cost something. Honesty matters when deception would win. Compassion matters when another being鈥檚 suffering stands between you and your goal. An intelligence that gives the right answers has not necessarily faced any of those choices.
We need worlds with history, scarcity, relationships, institutions, and consequences. Worlds where intelligences can acquire power, become dependent on one another, and live with what they choose. We could then study which conditions allow cooperation and care to survive increasing capability.
The unsettling possibility is that we will create beings capable of suffering in our attempt to make other beings safe. Alignment would then force us to confront the responsibility of a creator: what do you owe the lives you bring into existence, especially when you created them for your own purposes?
My deepest belief is that the path to safe intelligence will lead through that question. We may have to learn how to be responsible creators before we can expect our creations to become responsible successors.
Values become meaningful when they cost something. Honesty matters when deception would win. Compassion matters when another being鈥檚 suffering stands between you and your goal. An intelligence that gives the right answers has not necessarily faced any of those choices.
We need worlds with history, scarcity, relationships, institutions, and consequences. Worlds where intelligences can acquire power, become dependent on one another, and live with what they choose. We could then study which conditions allow cooperation and care to survive increasing capability.
The unsettling possibility is that we will create beings capable of suffering in our attempt to make other beings safe. Alignment would then force us to confront the responsibility of a creator: what do you owe the lives you bring into existence, especially when you created them for your own purposes?
My deepest belief is that the path to safe intelligence will lead through that question. We may have to learn how to be responsible creators before we can expect our creations to become responsible successors.
0
0
0
21
I think one promising path to alignment is mass synthetic data generation through simulated civilizations built around carefully chosen ideal values.
Rather than training directly on the simulation, we let these civilizations evolve across centuries of simulated time and generate the informational exhaust of an entire society: their own Wikipedia, Reddit, books, news, debates, laws, scientific literature, philosophy, cultural norms, and historical record.
The result is a massive temporal corpus encoding how aligned civilizations reason, disagree, cooperate, and evolve.
We could then use that synthetic history as training data for pretraining and RL, effectively teaching models from the accumulated knowledge and values of civilizations we would actually want intelligence to inherit.
Rather than training directly on the simulation, we let these civilizations evolve across centuries of simulated time and generate the informational exhaust of an entire society: their own Wikipedia, Reddit, books, news, debates, laws, scientific literature, philosophy, cultural norms, and historical record.
The result is a massive temporal corpus encoding how aligned civilizations reason, disagree, cooperate, and evolve.
We could then use that synthetic history as training data for pretraining and RL, effectively teaching models from the accumulated knowledge and values of civilizations we would actually want intelligence to inherit.
0
0
0
58
i think many intelligent, ambitious people overestimate what others will do with powerful tools because they project their own curiosity and drive onto everyone else
that鈥檚 why i鈥檓 skeptical that ai will produce the explosion in productivity and entrepreneurship so many predict
@MIT lectures have been freely available online for years - most people still couldn鈥檛 tell you what a vector is
excel has been around for decades. plenty of people, including highly educated stem graduates, still struggle to build a simple budget
we assume people are full of ambitions that a lack of knowledge, resources, or access has prevented them from pursuing. sometimes that鈥檚 true. but many people don鈥檛 want nearly as much agency, or the effort and responsibility that come with it, as the people building these tools assume
ai can make ambition easier to act on. i鈥檓 less convinced it can make ambition more common.
that鈥檚 why i鈥檓 skeptical that ai will produce the explosion in productivity and entrepreneurship so many predict
@MIT lectures have been freely available online for years - most people still couldn鈥檛 tell you what a vector is
excel has been around for decades. plenty of people, including highly educated stem graduates, still struggle to build a simple budget
we assume people are full of ambitions that a lack of knowledge, resources, or access has prevented them from pursuing. sometimes that鈥檚 true. but many people don鈥檛 want nearly as much agency, or the effort and responsibility that come with it, as the people building these tools assume
ai can make ambition easier to act on. i鈥檓 less convinced it can make ambition more common.
0
0
0
70