I think the primary gripe with AI is less about having no potential, but about the specific potential for evil being leveraged heavily while “visionaries” lie about or dismiss the (current and foreseeable future) limits for clout and profit.
To use your comparison, CEOs selling solar panels are responding to scepticism about the stability of that supply and its elasticity to fluctuating power demand with vague promises of developing more advanced cells that can operate with less light and throttle the generated power during ebbs in demand. They buy land to install more solar panels, then lobby for the deactivation of other, flexible power sources that could supplement their supply until they can finally start charging more in times of high demand.
They do so because the alternative would be admitting that the product they’re selling isn’t ready for the market they’re trying to enter (or at least not at the scale they’re aiming for) and that the solution to their issues requires different technologies which have not yet been developed. They can’t admit that because it would upset the investors that trusted their vision. As a famous adage suggests, they probably can’t even understand that limit, since their salary depends on not understanding it.
In the same vein, the current CEOs fail (or refuse?) to acknowledge that the generative models today have a fundamental limitation with understanding the semantics behind certain patterns. One infamous phenomenon would be AI hands: We know what a hand is and how it works; a generator can only vaguely imitate the patterns. It doesn’t know “hand” or “finger”, so it just generates some assembly that vaguely fits its training material.
If we consider the broader and ill-defined field of AI, I can believe that some form of semantic modelling technology could provide a way to connect tokens with abstract concepts and enable a type of reasoning that pure token predictors are incapable of. But we first need to acknowledge that limit, and on that front, we’re competing against tech-bros that would rather believe than know and against grifters that would rather sell snake oil than teach medicine.
I think the primary gripe with AI is less about having no potential, but about the specific potential for evil being leveraged heavily while “visionaries” lie about or dismiss the (current and foreseeable future) limits for clout and profit.
To use your comparison, CEOs selling solar panels are responding to scepticism about the stability of that supply and its elasticity to fluctuating power demand with vague promises of developing more advanced cells that can operate with less light and throttle the generated power during ebbs in demand. They buy land to install more solar panels, then lobby for the deactivation of other, flexible power sources that could supplement their supply until they can finally start charging more in times of high demand.
They do so because the alternative would be admitting that the product they’re selling isn’t ready for the market they’re trying to enter (or at least not at the scale they’re aiming for) and that the solution to their issues requires different technologies which have not yet been developed. They can’t admit that because it would upset the investors that trusted their vision. As a famous adage suggests, they probably can’t even understand that limit, since their salary depends on not understanding it.
In the same vein, the current CEOs fail (or refuse?) to acknowledge that the generative models today have a fundamental limitation with understanding the semantics behind certain patterns. One infamous phenomenon would be AI hands: We know what a hand is and how it works; a generator can only vaguely imitate the patterns. It doesn’t know “hand” or “finger”, so it just generates some assembly that vaguely fits its training material.
If we consider the broader and ill-defined field of AI, I can believe that some form of semantic modelling technology could provide a way to connect tokens with abstract concepts and enable a type of reasoning that pure token predictors are incapable of. But we first need to acknowledge that limit, and on that front, we’re competing against tech-bros that would rather believe than know and against grifters that would rather sell snake oil than teach medicine.