"Models are still expensive to run, hard to use, and frequently wrong".
The quote is true for human workers.
The quote is true for human workers.
We are having modern John Henry moment where the machines still stuck just enough that humans are ahead, but even small improvements in machines lead to a rapid replacements of parts of the workforce.
Much like we didn't stop improving drills, and no one in their right mind would drill a commercial tunnel without machine labor, our current AI models based in LLMs are still in their infancy, no one has any plans on stopping here.
We should also note that John Henry dies in the story from overworking, so we may want to take that in mind too.
People said the same stuff with the advent of computing. We’d all be living lives of leisure while computers did our jobs for us. It definitely didn’t work out that way.
Computers have in theory massively automated the world, but it also consolidates capital.
Salary isn't for delivering an untrustworthy result, its for being responsible for the result. AI doesn't take responsibility.
You've provided the answer yourself. The person who signs off on using the AI output bears the responsibility, unless the AI company provides some kind of guarantee, in which case the AI company bears responsibility.
These aren't new issues. That's why we have things like malpractice insurance, liability insurance, and so on.
>Prompt: Someone said "Models are still expensive to run, hard to use, and frequently wrong" Does this also apply to human worker?
>GPT4: This statement, while referring to AI models, can also apply to human workers in certain contexts. Here's a breakdown of each aspect mentioned:
> Expensive to run: Human workers may have higher costs associated with hiring, salaries, benefits, and overhead expenses. However, the costs of human workers can be justified in many cases due to their adaptability, creativity, and ability to learn.
> Hard to use: Human workers can be more difficult to manage compared to AI models, due to factors like communication, collaboration, and personal differences. However, humans are capable of adapting to new situations and understanding complex, nuanced scenarios.
> Frequently wrong: Humans, like AI models, are not infallible and can make mistakes or errors in judgment. However, humans possess the ability to learn from their mistakes, improve over time, and apply critical thinking to avoid making the same errors in the future.
> So, while there are similarities between the statement's implications for AI models and human workers, it is important to remember that both have their unique strengths and weaknesses. The best approach often involves combining the capabilities of AI and human workers to achieve optimal results.
On the other hand, the people drawing inferences from the exponential curve of Wikipedia's editor base growth about its potential to be a resource of unmatched accuracy were even more wrong, and it hit its quality ceiling pretty early...
Unlike wikipedia, with AI there is a threshold where it can start improving itself entering self improvement loop.
Similar to how computers can improve next generations of computers (with a lot of human effort) AI can improve AI at some point (with minimal or no human effort).
but that has pretty serious epistemological concerns. and to the people who are concerned about that - well i think that's funny... maybe take a walk down the science of how poor memory really is.
i guess i just have a very attenuated view of human (including my own) rationality. it's waaaaay more inflated for the average human than we're willing to admit.
i made a comment early on in December about the oncoming epistemological crisis and damn has it been really fun to watch unfold the last few months.
it feels like that would require much more sophistication, but then again who hasn't been amazed how intelligent ChatGPT & friends feel?
i would love to see some experiment where a debate is held/moderated by an LLM which can help find common ground in highly contentious viewpoints. the post-mortem from each side would be a fascinating read.