[AikiA] The Fifth Day of Christmas: Five Golden Rings
(media.kotakuinaction2.win)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (41)
sorted by:
LLMs are horrible at playing chess. Deep Blue and other chess "AIs" are fixed algorithmic programs that have little similarity to what we now know as AI/machine learning, which is based on context/statistics only.
Human errors are more predictable and they can also be held accountable. A human doing calculus will not suddenly decide that 2*2=8. An LLM can and does.
I agree they can get a lot worse, but probably not in the ways that people are predicting.
I wasn't talking about LLM's but AI's in general. There's no requirement for AI's to be stochastic or stochastic in a totally random way.
A good math AI wouldn't decide 2*2=8 either. A human being could mistakenly use an 8 however.
All the hype over AI restructuring the world in the last 5 years is based on LLMs.
OK but future AI's probably won't be LLM's, they are just a stepping stone. A good general purpose AI would be unpredictable in some areas and predictable in others (like math). LLM's are already somewhat predictable when they need to be but could be improved.
I'll wait and see. I assumed that kind of thinking would have made it into AIs already.