This needs to be repeated loudly and often.
I hear fellow tech folks all the time say things like, "LLMs actually understand this kind of data".
Instead: "LLMs generate grammatical statements more or less related to this kind of data."
I hear fellow tech folks all the time say things like, "LLMs actually understand this kind of data".
Instead: "LLMs generate grammatical statements more or less related to this kind of data."
Reposted from
🏳️⚧️ Alexandra Merideth Erin [she/her]
An AI does not misunderstand anything because an AI does not operate on the basis of understanding.
Human minds can understand something, so we are capable of misunderstanding something.
An LLM that generates a statement has performed its function correctly whether the statement is true or false.
Human minds can understand something, so we are capable of misunderstanding something.
An LLM that generates a statement has performed its function correctly whether the statement is true or false.
Comments