Post #3885906

2026-07-17 11:46 UTC

@geraldew@fosstodon.org also the bit where if you ask for the reasoning chain it doesn't give you the one it used, it synthesizes something that sounds like a reasoning chain. This is apparently for security and not from incompetence

Replies (1)

@abucci@buc.ci 2026-07-17 13:21

@davidgerard@circumstances.run @geraldew@fosstodon.org These guys seem to be constitutionally unable to think outside the neural network. If one neural network is misbehaving the go-to answer is to glom another neural network onto it somehow. There's an ideology behind it ("connectionism"). We have perfectly functional logical reasoning engines that can accurately self-report, and have had these for decades by this point. One way to understand some of these AI systems is as free riders: they use some of the language of proper reasoning engines to give the impression their mess of neural network Play-Doh is the real logical deal when it is not. A proper hybrid might be, as Gary Marcus has argued for instance, but that would run afoul of the "always use neural networks" dictate.

Open ##3885902