If AI ends up running companies better than people, won’t shareholders demand the switch? A board isn’t paying a CEO $20 million a year for tradition, they’re paying for results. If an AI can do the job cheaper and get better returns, investors will force it.
And since corporations are already treated as “people” under the law, replacing a human CEO with an AI isn’t just swapping a worker for a machine, it’s one “person” handing control to another.
That means CEOs would eventually have to replace themselves, not because they want to, but because the system leaves them no choice. And AI would be considered a “person” under the law.
It’ll take a few years but it progresses exponentially, it will get there.
It progresses logistically; eventually it’ll plateau and there’s no reason to believe that plateau will come after “can do everything a human can.”. See: https://www.promptlayer.com/research-papers/have-llms-hit-their-limit
Sure, but we don’t know where that plateau will come and until we get close to it progress looks approximately exponential.
We do know that it’s possible for AI to reach at least human levels of capability, because we have an existence proof (humans themselves). Whether stuff based off of LLMs will get there without some sort of additional new revolutionary components, we can’t tell yet. We won’t know until we actually hit that plateau.
Current Ai has no shot of being as smart as humans, it’s simply not sophisticated enough.
And that’s not to say that current llms aren’t impressive, they are, but the human brain is just on a whole different level.
And just to think about on a base level, LLM inference can run off a few gpus, roughly order of 100 billion transistors. That’s roughly on par with the number of neurons, but each neuron has an average of 10,000 connections, that are capable of or rewiring themselves to new neurons.
And there are so many distinct types of neurons, with over 10,000 unique proteins.
On top of there over a hundred neurotransmitters, and we’re not even sure we’ve identified them all.
And all of that is still connected to a system that integrates all of our senses, while current AI is pure text, with separate parts bolted onto it for other things.
The human brain is doing a lot of stuff that’s completely unrelated to “being intelligent.” It’s running a big messy body, it’s supporting its own biological activity, it’s running immune system operations for itself, and so forth. You can’t directly compare their complexity like this.
It turns out that some of the thinky things that humans did with their brains that we assumed were hugely complicated could be replicated on a commodity GPU with just a couple of gigabytes of memory. I don’t think it’s safe to assume that everything else we do is as complicated as we thought either.
Yeah a lot of it is messy, but they are not being replicated by commodity gpus.
LLMs have no intelligence. They are just exceedingly well at language, which has a lot of human knowledge in it. Just read claudes system prompt and tell me it’s still smart, when it needs to be told 4 separate times to avoid copyright.
Hm… two bucks… and it only transports matter? Hm…
It’s amazing how quickly people dismiss technological capabilities as mundane that would have been miraculous just a few years earlier.
I did not immediately dismiss LLM, my thoughts come from experience, observing the pace of improvement, and investigating how and why LLMs work.
They do not think, they simply execute an algorithm. Yeah that algorithm is exceedingly large and complicated, but there’s still no thought, there’s no learning outside of training. Unlike humans who are always learning, even if they don’t look like it, and our brains are constantly rewiring themselves, LLMs don’t.
I’m certain in the future we will get true AI, but it’s not here yet.
Ive had too many beers to read that.