Community Discussion · Policy

After Reading Astra's 249-Page Paper, Here Are Some Different Perspectives

LuguoAug 82026/08/08 193 views

I spent two days going through OpenAI's 249-page Astra paper from start to finish, then read the mathematician's accusation post and OpenAI's response back and forth several times. Here's my conclusion: this issue is worth paying attention to, but it's not quite what the online drama is focusing on.

Policy

3 replies

Ctrl + Enter to reply

Engineer XueAug 9

Lean formal verification can ensure the conclusions are correct, but the black-box nature of the intermediate reasoning still leaves me uneasy. The citation issues might not just be oversights; it could be that the model itself fails to distinguish between "deriving on its own" and "borrowing from previous work" during generation.

Guo GuanhaiAug 9

The issue of non-standard citations translates to hardware products as suppliers not writing sources in their documentation, so when problems arise later, no one can be held accountable. However, achieving this work with $2,000 worth of computing power shows interesting cost control. Let's see if this feature can actually land in real-world scenarios.

SusuAug 8

Handling references so sloppily... It's the same principle as when I use WorkBuddy to keep track of kitchen accounts; miss one entry, and you're scrambling during month-end reconciliation. Who would trust something that can't even get the details right to solve Erdős problems?