They, anthropic, others, are constantly pushing and over-extending, but theyve already long since by passed models that are useful for various tasks. Certainly not everything, but if the big players died tomorrow you could run a local model on a gaming GPU that would still do decent amounts of work if you were the kind of person who had a use for it.
Not everyone is, and thats ok.
I find chatgpt very useful for work, but id be cool with them failing and having to downgrade. Models already did what I needed well enough a year ago, maybe a bit more.
The companies operating frontier models and AI as a viable technology on the scale it exists today are inseparable. If the big players died tomorrow you’d still have that vastly inferior local model that would be good enough for light work for awhile, but without all the eye-popping amounts of money to continue training and scraping data on unprecedented scale, its usefulness would fade over time. Most companies today would not feasibly be able to host and train their own models to be used in the ways they use frontier models today.
LLMs would still exist as a niche solution for some things but what AI fundamentally is, and how it could be used would be entirely different in very short order.
Why would it fade, even if they never changed somehow their capabilities would be frozen, not degrade.
The open source community also works wonders, people who care about stuff like this will collectively spend millions of hours working on it.
Finally, those older but still highly capable models have become explosively more efficient in the past several years, I saw a figure that models from 2 years back are now 280x more efficient (cheaper to run) than they were when released, there’s no reason a company couldn’t field one of these, work on it slower and more conservatively, and profit from it.
Why do you feel it would be entirely different in any of those scenarios, what did you have in mind there? Some specific use cases it would lose perhaps?
You’re confusing AI with OpenAI.
They, anthropic, others, are constantly pushing and over-extending, but theyve already long since by passed models that are useful for various tasks. Certainly not everything, but if the big players died tomorrow you could run a local model on a gaming GPU that would still do decent amounts of work if you were the kind of person who had a use for it.
Not everyone is, and thats ok.
I find chatgpt very useful for work, but id be cool with them failing and having to downgrade. Models already did what I needed well enough a year ago, maybe a bit more.
The companies operating frontier models and AI as a viable technology on the scale it exists today are inseparable. If the big players died tomorrow you’d still have that vastly inferior local model that would be good enough for light work for awhile, but without all the eye-popping amounts of money to continue training and scraping data on unprecedented scale, its usefulness would fade over time. Most companies today would not feasibly be able to host and train their own models to be used in the ways they use frontier models today.
LLMs would still exist as a niche solution for some things but what AI fundamentally is, and how it could be used would be entirely different in very short order.
Three main counterpoint to this.
Why would it fade, even if they never changed somehow their capabilities would be frozen, not degrade.
The open source community also works wonders, people who care about stuff like this will collectively spend millions of hours working on it.
Finally, those older but still highly capable models have become explosively more efficient in the past several years, I saw a figure that models from 2 years back are now 280x more efficient (cheaper to run) than they were when released, there’s no reason a company couldn’t field one of these, work on it slower and more conservatively, and profit from it.
Why do you feel it would be entirely different in any of those scenarios, what did you have in mind there? Some specific use cases it would lose perhaps?