←back to thread

419 points serjester | 2 comments | | HN request time: 0.588s | source
Show context
marban ◴[] No.43536553[source]
Giving up accuracy for a bit of convenience—if any at all—almost never pays off. Looking at you, Alexa.
replies(1): >>43536981 #
1. danielbln ◴[] No.43536981[source]
Image compression, eventual consistency, fuzzy search. There are many more examples I'm sure.
replies(1): >>43537161 #
2. skydhash ◴[] No.43537161[source]
> Image compression, eventual consistency, fuzzy search. There are many more examples I'm sure.

Isn't all of these very deterministic? You can predict what's going to be discarded by the compression algorithm. Eventual consistency is only eventual because of the generation of events. Once that stops, you will have a consistent system and the whole thing can be replayed based on the history of events. Even with fuzzy search you can intuit how to get reliable results and ordering without even looking at the algorithms.

An LLMs based agent is the least efficient method for most of the cases they're marketing if for. Sometimes all you need is a rule-based engine. Then you can add bounded fuzziness where it's actually helpful.