4 ms·
In classical terms, out of pathos, logos and ethos, LLMs are logos. They discover metrics. https://en.wikipedia.org/wiki/Metric_space https://en.wikipedia.org/
by ganzuul 3y ago
In classical terms, out of pathos, logos and ethos, LLMs are logos. They discover metrics.
https://en.wikipedia.org/wiki/Metric_space https://en.wikipedia.org/wiki/Metric_space
Neural networks are an optimization which exploits structure in sparsity.
Intelligence is not a black box.
- justinl33 3y agoI agree that the mathematics motivating neural networks are well understood. We understand the heuristics behind them; obviously, we created them and they work. I more just mean at the most fundamental level, do we understand intelligence; could we reconstruct it without just ‘optimizing’ towards it. If we understand intelligence then we should be able to deterministically explain all behaviour of neural networks/LLM’s, instead of just poking them from the outside and seeing how they react.
- ganzuul 3y agoI don't think poking them will lead to much understanding of intelligence.
- white_dragon88 3y ago[dead]