Post #3699547
2026-07-06 03:29 UTC
@BenjaminHCCarr@hachyderm.io Before the LLM craze took over, we actually had fairly decent speech to text, though not perfect. Imagine where we might be now if all that "AI" money had focused on improving useful deterministic programs like that, instead of on "how can we trick and deceive people on the largest scale possible".
Replies (1)
-
@harisont@mstdn.social 2026-07-06 08:34
@lauerhahn@sfba.social @BenjaminHCCarr@hachyderm.io I am way more impressed with and positive about STT models than LLMs in this context, but speech recognition (that works well) is also not deterministic (at least not in the sense in which LLMs aren't). They're specialized (and therefore more predictable) machine learning models though, which I much prefer over these unreliable factotums