TypeSafe AI says reliable decisions require combining reliable judgments with code

Nov 6

1. 09:54

   TypeSafe AIFirst-party

   TypeSafe AI: Data Distributions Are More Absurd Than Your Tricks

   TypeSafe AI argues that hiding semantic logic inside string prompts, embedding vectors, or complex ML is useless—data distributions are always much smarter than people. The author suggests: clearly define what you need to know, and confirm that reading the data directly is enough to know it, so you can make reliable decisions.

   Expand summary
2. 09:54

   TypeSafe AIFirst-party

   TypeSafe AI: Reliable decisions rely on code, not decision models

   TypeSafe AI stated in a tweet that reliable decisions cannot depend on decision models: hiding semantic logic in string prompts, embedding vectors, or complex ML is useless—data distributions are always smarter than people. Its argument is to clearly identify what needs to be known and confirm that reading the data directly is sufficient, then use code to compose reliable judgments, and it claims that Jev can make reliable judgments.

   Expand summary
3. 09:54

   TypeSafe AIFirst-party

   TypeSafe AI: Reliable decisions rely on code, not decision models

   TypeSafe AI responds to "why not call Jev a decision model": when faced with complex data, the model itself cannot guarantee reliable decisions; reliability requires code, and reliable decisions come from composing reliable judgments in code. It adds that you can hide semantic logic in string additional conditions, embedding vectors, and various ML, but data distributions are always stranger than you can imagine, so you must be clear about what you need to know and confirm that reading the data directly is sufficient.

   Expand summary