A company is trying to accurately state not only 'what it can do' but also 'what it cannot do' in a generative AI adoption proposal. Which TWO of the following are appropriate as inherent limitations of generative AI? (Choose TWO.)
Select two inherent limitations of generative AI.
It does not know the latest information after the training data cutoff.
This is correct. The model only has knowledge up to the point it was trained (the cutoff), so it does not know the latest information after that. The latest external information must be supplemented with RAG and the like.
It can make mistakes in precise calculation or logical reasoning.
This is correct. Because generative AI generates words probabilistically, it can make mistakes in precise calculation or multi-step logical reasoning. If accuracy is needed, supplement with verification or dedicated tools.
Compute cost is incurred each time inference is run.
It is true that compute cost is incurred for each inference, but this is an operational cost characteristic, not something the model fundamentally 'cannot do' (an inherent limitation).
It is not a capability limitation that this question asks about, so it is incorrect.
There is latency in generating a response.
It is true that some latency occurs before a response, but this is a performance/operational characteristic, not something the model fundamentally cannot do (an inherent limitation).
It is not a capability limitation that this question asks about, so it is incorrect.
It does not work at all without a cloud environment.
There are foundation models that run on-premises or on edge devices, so the claim that the cloud is mandatory is factually incorrect.
It is neither an inherent limitation nor a fact, so it is incorrect.
Inherent limitations of generative AI include 'not knowing the latest information after the training data cutoff' (supplement with RAG and the like) and 'being able to make mistakes in precise calculation or logical reasoning' (supplement with verification or dedicated tools).