This is what I started thinking last night. Any of the people who were added to the chat could be disgruntled and add a reporter to the chat to leak it. Is there even any log of who added who to a chat? There might not even be any way to pin it on the leaker. If the leaker had been involved in several such chats and knew the intent was to intentionally violate federal recordkeeping laws, not only would this accomplish the leak, but there might not even be any record of who caused the leak.

Run r1 671b at a q6 quant?

Don't they have a standalone Grok app now? I thought I saw that. [edit] ah some sibling comments mention this as well

Training a model to reason that it should write and execute code is the solution here. Synthetic datasets showing lots of math questions followed by reasoning that writes and executes the program for the answer.

Waffle machines is way better. Love it. Thanks.