A calculator app is also incapable of working with letters, does that show that the calculator is not reliable?
What it shows, badly, is that LLMs offer confident answers in situations where their answers are likely wrong. But it’d be much better to show that with examples that aren’t based on inherent technological limitations.
“It”? Are you conflating the low parameter model that Google uses to generate quick answers with every AI model?
Yes, Google’s quick answer product is largely useless. This is because it’s a cheap model. Google serves billions of searches per day and isn’t going to be paying premium prices to use high parameter models.
You get what you pay for, and nobody pays for Google so their product produces the cheapest possible results and, unsurprisingly, cheap AI models are more prone to error.
You aren’t wrong about why it happens, but that’s irrelevant to the end user.
The result is that it can give some hilariously incorrect responses at times, and therefore it’s not a reliable means of information.
A calculator app is also incapable of working with letters, does that show that the calculator is not reliable?
What it shows, badly, is that LLMs offer confident answers in situations where their answers are likely wrong. But it’d be much better to show that with examples that aren’t based on inherent technological limitations.
“It”? Are you conflating the low parameter model that Google uses to generate quick answers with every AI model?
Yes, Google’s quick answer product is largely useless. This is because it’s a cheap model. Google serves billions of searches per day and isn’t going to be paying premium prices to use high parameter models.
You get what you pay for, and nobody pays for Google so their product produces the cheapest possible results and, unsurprisingly, cheap AI models are more prone to error.