> future models are world-changing
Possibly, but talking about what you imagine the future will be in the present tense is hype. That's not the future, it's speculation about one scenario. There are others.
There's been a lot of progress making models that aren't quite as good as GPT-4, or are maybe comparable. Why is that? Have we hit a wall? I don't know, but that's a scenario, too.
So how do you know this exactly?
What the naysayers always miss is exponential growth. They don’t look towards the future.
This is quite a pat response. There are plenty of critics within the AI/ML field of the unbounded growth hypothesis that you are effectively fronting here. I don't think you really believe that they "don't look towards the future", or that they aren't aware of the concept of exponential growth.
I'd be interested in a cohesive response to their arguments. But in the above post at least, I'm not seeing one.
Exponential growth isn’t _normal_, tho. I think that Moore’s Law has messed us up a bit, honestly; there is a tendency to expect _everything_ to behave that way. And so far, with LLMs, it just doesn’t seem to be happening. When you double the compute, you double the amazingness of the LLM, right? Well, actually, no; based on current results there’s little reason to think that, and if that doesn’t hold your exponential growth is in _serious_ doubt.
> but also because the market is anticipating huge advancements in model ability
Markets anticipate all sorts of things which never, in the end, happen, remember.
> Current models are limited, but future models are world-changing
See, this is why people draw the comparison to blockchain stuff. “Yeah, the current thing’s a bit shit, but just wait for the next one. In the meantime, can I convince you to invest a hundred million in my robot tax advisor?”
> and we have every reason to believe models will continue to scale in this way.
Do we? I mean, to me it looks like it has clearly slowed down, already.
Is current-gen AI valuable? Obviously. Look at how many people say as much and willingly pay for the services.
Is current-gen AI worth the hype? No, not really. Current models are impressive but highly limited, as any user can attest (and as any hater will focus on).
What the naysayers always miss is exponential growth. They don’t look towards the future. Nvidia’s stock is high because of hype, yes, but also because the market is anticipating huge advancements in model ability. Current models are limited, but future models are world-changing, and we have every reason to believe models will continue to scale in this way.
Just you wait for GPT-5.