HN Explorer

Ask HN: Are LLMs just the right-brain?

1 points by stevenalowe ↗ HN
And if so, does that make us (via prompting) a left-brain?

LLMs are fantastic at generating believable nonsense, a creative stream of expression that superficially resembles reality. This is not a criticism; what LLms can do is a superpower. But the left-brain of AGI is still MIA. Are model builders aiming to fill that void, or do we need a different mechanism? If so, what might that be?

4 comments

[ 5.3 ms ] story [ 35.2 ms ] thread
incomingpain
The "left brain" vs "right brain" thing is a myth for humans.

LLMs are more than left vs right brained. They have multiple experts; they have multiple modality.

Some big models can have hundreds of "side of brain"

stevenalowe
Of course, using left/right brain as an analogy. Read it as A/B brain pairs like Minsky’s society of mind if you prefers. Bag of experts under the hood doesn’t matter if the whole spews nonsense. Right now, we check the output, the model does not check itself
namegulf
From a technical perspective, they're doing exactly what they're designed to do:

Self learn (i.e. unsupervised) patterns (using multi-head attention mechanism) from a large volume of input and then generate new patterns based on the. prompt.

stevenalowe
Yes I know how they work, the question is why they can’t self-evaluate