One day an AI might be, it won’t be an LLM though.
I’m no expert, but I feel like that’s akin to saying a human may be intelligent, but it won’t be Broca’s area of the brain.
I would think a true AGI would involve a lot of different neural networks doing different things. And an LLM could end up being a useful component of allowing it to communicate. Though who knows how the model would need to change if its language prompts are inputs from other neural nets.
And I have to imagine at this point major AI models aren’t just LLMs, that’s just the part we interface with.
Again, I’m very far from an expert. What I do know is that the current MASSIVE push with the technology* we have now is causing more harm than it’s worth.
Brocas area on its own isn’t intelligent and strapping a bunch of Brocas areas together won’t get you there either.
Most of the frontier models as far as I’m aware are basically a bunch of differently trained LLMs strapped together and even then there have only been incremental improvements to their performance, no new functionality has really emerged from doing that.
LLM investment is IMO a dead end hype train and will require breakthroughs in other techniques of machine learning to put together something we would recognise as truly intelligent. I’ll concede the possibility that LLM like functionality may be a portion of that but equally it may not be.
I’m no expert, but I feel like that’s akin to saying a human may be intelligent, but it won’t be Broca’s area of the brain.
I would think a true AGI would involve a lot of different neural networks doing different things. And an LLM could end up being a useful component of allowing it to communicate. Though who knows how the model would need to change if its language prompts are inputs from other neural nets.
And I have to imagine at this point major AI models aren’t just LLMs, that’s just the part we interface with.
Again, I’m very far from an expert. What I do know is that the current MASSIVE push with the technology* we have now is causing more harm than it’s worth.
*edit: typo
Brocas area on its own isn’t intelligent and strapping a bunch of Brocas areas together won’t get you there either.
Most of the frontier models as far as I’m aware are basically a bunch of differently trained LLMs strapped together and even then there have only been incremental improvements to their performance, no new functionality has really emerged from doing that.
LLM investment is IMO a dead end hype train and will require breakthroughs in other techniques of machine learning to put together something we would recognise as truly intelligent. I’ll concede the possibility that LLM like functionality may be a portion of that but equally it may not be.