Generative AI is useful, in the same way that electricity is useful.
I use Claude for a wide range of tasks, every day.
And today is the worst that Claude will ever be. Claude – and other generative AI services – will never ever again be as crap as it is today.
- Claude is inconvenient. It’s a web-based chatbot that isn’t integrated into any of the services I use. There’s a lot of copying and pasting going on.
- Claude isn’t multimodal. I can’t use it with audio. Or video. Or images. I can’t talk to it, and it can’t draw for me, and it can’t summarise videos that I don’t have time to watch.
- Claude is expensive. It’s expensive given that I expect it to be like wifi, which is essentially free given how much of that I consume.
- Claude hallucinates. No matter how big the training dataset is, or how many GPU clusters are added to the process, Claude will never be grounded in reality. We may approximate reality but the nature of LLM technology seems to suggest that it can’t reason.
- Claude is biased. Because of the way that human beings are biased, and we create the data used to train LLMs, LLMs will likely be biased for a long time.
- Claude is not transparent. We have no real idea how Claude was trained, and on what data.
- Claude is a black box. We can’t ask Claude to explain it’s reasoning (even though it responds as if it is reasoning) because it isn’t reasoning. And it may be the case that it can never reason.
Make no mistake, generative AI is inconvenient, expensive, disconnected from reality, biased, opaque, and can’t explain itself.
But even with all these problems, I still use Claude every day. Every day it helps me to think through complex problems. Every day it gives me creative ideas. Every day it provides me with context. Every day it points out my blind spots. Every. Single. Day.
This early stage of generative AI has deep problems, in the same way that the early stages of electricity production and distribution had deep problems. But, just like most of the problems with electricity were ironed out, we’ll iron out many of the problems with generative AI.
Tomorrow Claude will be cheaper, more convenient, multimodal, will hallucinate less, and be less biased. We’ll have more regulation about how data is used to train foundation models. On it’s own it may never be able to reason, but integration with other AI paradigms will address this. However, even if none of this comes to pass and today’s generative AI is the best we’ll ever have, it still represents a fundamental shift in society.
Because it is useful. In the same way that electricity is useful.