3 ms·

I think it does represent a dead end, but not for the reasons presented in this article. The real issue in my opinion is that we will hit practical limits with

I think it does represent a dead end, but not for the reasons presented in this article. The real issue in my opinion is that we will hit practical limits with training data and computational resources well before AGI turns us all into paperclips, basically there is no "Moore's Law" for AI and we are already slowing down using existing models like GPT. We are in the vertical scaling phase of AI model development, which is not sustainable long-term. I discussed this further here for anyone interested: https://techleader.pro/a/658-There-is-no-Moore's-Law-for-AI-(TLP-2024w35) https://techleader.pro/a/658-There-is-no-Moore's-Law-for-AI-...

EerkeBoiten 2y ago
Author here. Fair enough. The "dead end" in the title isn't mine, extracted by editor from what's essentially a side comment on the limits of LLM scaling. My title would have been something like "We can't use current AI for critical applications". With "for" being essentially different from "in", if you want the nuance there.
NoGravitas 2y ago
> The real issue in my opinion is that we will hit practical limits with training data and computational resources well before AGI turns us all into paperclips [...] I think you are correct, but also I think that even if that were not the case, the Thai Library Problem[1] strongly suggests that AGI will have to be built on something other than LLMs (even if LLM-derived systems were to serve as an interface to such systems). [1]: https://medium.com/@emilymenonbender/thought-experiment-in-the-national-library-of-thailand-f2bf761a8a83 https://medium.com/@emilymenonbender/thought-experiment-in-t...