Maybe this is just a contrarian view, but I see “AI” as a potential rather than a technology. Right now, transformer-based technologies are what most of us mean when we talk about AI, and it’s not clear to me how much more potential that idea really has. When I look at how much energy it takes to set up something like GPT-4 I see us pushing hardware to its limit and yet the outcomes are still too often unsatisfying. Significant breakthroughs are needed somewhere in that architecture just to do the kind of things we’re trying to do today at the fidelity we expect and without breaking the bank.
The technology we have today might be to AI what the phonograph was to audio recording. As a technology we hit the limits of its potential pretty quickly and then… we fixated. Entirely different technologies eventually led to the lossless spatial audio experiences we can enjoy today, and seem more likely to carry future potential for audio too.
In that analogy, GPT might just be like someone arranging 8 gramophones in a circle to mimic the kind of spatial audio experience available in some headphones now. Impressive in many ways, but directionally not the path where potential lies.
I agree. Many people are fixated on GPT because it is shiny and novel, but it is certainly not the pinnacle of what AI could be, or even close. One day, we will look back on calling GPT an “AI” like we would someone calling the first two tin cans on a string a “phone”. Accurate enough, but certainly a far cry from any modern phone.
AI has the capacity to be the most impactful overall to our daily lives, but like most things, advancement will continue to be limited by hardware.
Maybe this is just a contrarian view, but I see “AI” as a potential rather than a technology. Right now, transformer-based technologies are what most of us mean when we talk about AI, and it’s not clear to me how much more potential that idea really has. When I look at how much energy it takes to set up something like GPT-4 I see us pushing hardware to its limit and yet the outcomes are still too often unsatisfying. Significant breakthroughs are needed somewhere in that architecture just to do the kind of things we’re trying to do today at the fidelity we expect and without breaking the bank.
The technology we have today might be to AI what the phonograph was to audio recording. As a technology we hit the limits of its potential pretty quickly and then… we fixated. Entirely different technologies eventually led to the lossless spatial audio experiences we can enjoy today, and seem more likely to carry future potential for audio too.
In that analogy, GPT might just be like someone arranging 8 gramophones in a circle to mimic the kind of spatial audio experience available in some headphones now. Impressive in many ways, but directionally not the path where potential lies.
I agree. Many people are fixated on GPT because it is shiny and novel, but it is certainly not the pinnacle of what AI could be, or even close. One day, we will look back on calling GPT an “AI” like we would someone calling the first two tin cans on a string a “phone”. Accurate enough, but certainly a far cry from any modern phone.
AI has the capacity to be the most impactful overall to our daily lives, but like most things, advancement will continue to be limited by hardware.