Okay, folks, the world’s starting to realize that AI is going to automate a lot of jobs faster than we thought. But here’s the kicker: we’re still not really talking about who gets the money and power from all this.
The Big Debate: Robots Taking Our Jobs?
Just today, the Vice President said AI will just boost productivity, not replace workers. But then you’ve got Sam Altman, the head honcho at OpenAI, saying that labor might lose out to capital. And the RAND Corporation, a well-known think tank, put out a paper saying we’re not ready for the job losses and social unrest that might come with super-smart AI.
Even if companies do get richer, who decides how the money gets split? Elon Musk is even challenging Sam Altman and Microsoft for control of OpenAI. It’s like a high-stakes game of chess!
AGI for $20? Seriously?
And get this: some folks at Stanford are saying that you can make AI way smarter for just $20. That’s less than your Netflix subscription!
Dario Amodei, the CEO of Anthropic, which created Claude, warns that time is running out to control AGI.
When AI can do almost anything, let’s hope we’re more united than we are now. There’s a lot happening, so let’s break down the most interesting stuff, using Sam Altman’s essay as a starting point.
What Exactly IS “AGI” Anyway?
Sam Altman is throwing around yet another definition of AGI. This time, it’s a system that can solve tough problems at a human level in many areas. And guess what? We’re getting close.
AI Can Code Now?!
Take coding, for example. OpenAI’s models are getting seriously good. The O3 model was once ranked among the top 175 coders in Codeforces ELO. Now, they have an internal model scoring within the top 50! According to Sam Alman in Japan. These systems aren’t just copying top coders; they’re figuring things out themselves using reinforcement learning. They’re teaching themselves what works.
AI Doctors? Maybe Soon!
I’ve even been using OpenAI’s deep research to suggest diagnoses for a relative, and a doctor friend said it found things she wouldn’t have thought of. Sure, it hallucinates sometimes, but it also comes up with new ideas. And that’s just with the older models searching a limited number of sources! Imagine what the newer ones could do.
No Limit to What AI Can Learn
Karina Nenn from OpenAI says they’re crushing all the benchmarks. They’ve gone from just feeding data to AI to teaching it an infinite number of tasks through reinforcement learning. How to search the web, use a computer, write well – you name it. The benchmark saturation is because of reinforcment learning
She thinks the bottleneck now is actually evaluating AI, not training it.
Even though their current system is a bit clunky, tasks like buying stuff online or filling out spreadsheets are easily verifiable. And when something is verifiable, AI can master it through reinforcement learning.
The Money Pit: How Much Will AI Cost?
Sam Altman says that the intelligence of an AI model is roughly equal to the log of the resources used to train and run it. In other words, you have to keep pouring in resources to make it smarter.
But here’s the crazy part: even a small increase in intelligence can be super valuable. If someone doubled the intelligence of one of these models, it would be worth way more than double the investment.
That’s why Altman sees no reason for the exponentially increasing investment to stop anytime soon. If AI pays you back tenfold, why would you ever stop investing?
OpenAI’s Grand Plan: World Domination (and Redistribution?)
Less than two years ago, Altman said his crazy idea was that OpenAI would grab a huge chunk of the world’s wealth through AGI and then give it back to the people. We’re talking trillions of dollars!
Of course, he’s not sure how they’d redistribute it. But that’s the scale we’re talking about – the scale of the entire labor force of the planet.
The Battle for Control
Unsurprisingly, not everyone wants Altman to have that much power. Elon Musk reportedly bid almost $100 billion for OpenAI.
It looks like Altman and OpenAI have valued the nonprofit stake in OpenAI at around $40 billion. But if Musk and others have valued that stake at $100 billion, it might be hard for Altman to argue in court that it’s worth only $40 billion.
Altman told employees that these are just tactics to weaken them. The nonprofit behind OpenAI could also reject the offer because it thinks AGI wouldn’t be safe in Musk’s hands.
Cheaper TVs, Crazy Expensive Land?
Altman predicts that with AGI, the price of many goods will fall dramatically. That might make up for people losing their jobs. But he also says the price of luxury goods and land may rise even more dramatically.
He might have one particular luxury good in mind: a hardware device designed by Johnny Ive from Apple. Altman says it’s incredible and just a year away.
On the other hand, smaller language models might become more accessible. Altman mentioned the idea of open-sourcing smaller versions of their models.
AI for All… Really?
The mission of OpenAI is to ensure that AGI benefits all of humanity, not just most of humanity. But how are they going to achieve that when they admit that most human labor might become redundant?
Yoshua Bengio, a leading AI researcher, thinks that if one nation gets to AGI before another, they might use that advantage to wipe out the economies of other nations.
He also thinks that the companies that control these AI systems will keep the really powerful ones for themselves and use them to compete with existing companies.
Google’s Response: Gemini and More
Google is fighting back with its Gemini models. While their benchmark results aren’t quite as good as OpenAI’s, they’re decent.
Gemini is amazing at quickly reading vast amounts of PDFs and other files, and it’s incredibly cheap.
As ChatGPT becomes more popular, Google will invest more and more to make sure that Gemini is state-of-the-art.
The Dark Side: AI and Authoritarianism
Altman wrote that AI could be used by authoritarian governments to control their population through mass surveillance and loss of autonomy.
The RAND paper also warns about other threats to national security, such as wonder weapons, systemic shifts in power, non-experts empowered to develop weapons of mass destruction, and artificial entities with agency.
The paper admits that the U.S. is not well-positioned to realize the economic benefits of AGI without widespread unemployment and societal unrest.
AGI for $20? The Sequel!
For less than $50 worth of compute time of course not counting the reasearch time , Stanford produced S1 now yes of course they did utilize an openweight base model quen 2.5 32 billion parameters instruct but the headline is with just a th000 questions worth of data they could bring that tiny model to being competitive with 01
They used an open-weight base model and just a thousand questions worth of data. The key was to force the model to keep going when it wanted to stop.
This shows that even small models can achieve impressive results with the right techniques.
The Future is Uncertain
So, what does all this mean? The truth is, nobody knows for sure. But one thing is clear: AI is going to change everything. The question is, will it be for the better?