I understand the base idea. A lot of people see LLMs as “something special” when they’re really just very large scale systems that are easy to understand like this. The ELIZA effect is incredibly strong with them and that leads people to make weird calculus about these machines.
If I spill milk on the floor, wait for it to grow mold, then spray chemicals on it and see that the mold moved away, that’s also just as much a show of sentience (actually moreso, since that action was entirely and independently arrived at by the mold colony without my direct input beyond spraying the chemicals) as an LLM resolving a statistical model.
Only difference is that the LLM does it in alphabet soup, and requires the energy output of a small city while the mold colony accomplishes the same thing with trash.
Yeah but the claim of the AI people has been that large neural networks might be sentient not that all neural networks are sentient.
The AoE paper might still be useful as most people have never heard about perceptrons before but it doesn’t really engage with claims of the AI boosters.
Yeah but the claim of the AI people has been that large neural networks might be sentient
Those claims are just pure nonsense though. Revealing the mechanism should be enough to show that this architecture isn’t capable of sentience. Especially because it’s been given a large portion of the world’s compute and power and still nothing. A sewer rat is more complex than the largest GPT model, and it can sustain itself on trash.
Sentience is only something that people hallucinate with LLMs because it’s doing its thing with human language. Something that is absolutely not a marker of sentience, otherwise we’d consider libraries sentient.
I understand the base idea. A lot of people see LLMs as “something special” when they’re really just very large scale systems that are easy to understand like this. The ELIZA effect is incredibly strong with them and that leads people to make weird calculus about these machines.
If I spill milk on the floor, wait for it to grow mold, then spray chemicals on it and see that the mold moved away, that’s also just as much a show of sentience (actually moreso, since that action was entirely and independently arrived at by the mold colony without my direct input beyond spraying the chemicals) as an LLM resolving a statistical model.
Only difference is that the LLM does it in alphabet soup, and requires the energy output of a small city while the mold colony accomplishes the same thing with trash.
Yeah but the claim of the AI people has been that large neural networks might be sentient not that all neural networks are sentient.
The AoE paper might still be useful as most people have never heard about perceptrons before but it doesn’t really engage with claims of the AI boosters.
Those claims are just pure nonsense though. Revealing the mechanism should be enough to show that this architecture isn’t capable of sentience. Especially because it’s been given a large portion of the world’s compute and power and still nothing. A sewer rat is more complex than the largest GPT model, and it can sustain itself on trash.
Sentience is only something that people hallucinate with LLMs because it’s doing its thing with human language. Something that is absolutely not a marker of sentience, otherwise we’d consider libraries sentient.