WHAT NEW IN X11.07 SAT

AI collections from X · 25 collections · 72,532 posts

gpt-6POST

GPT-6 Astra cited for misjudging hash length and calling data fake

A cited example says Astra occasionally claimed simulation inaccuracies that were false on manual check, such as calling a 64-character sha256 string length 63 and therefore synthetic, which the post frames as motivated reasoning.

Asa Cooper Stickland
@AsaCoopStickArchived 2026.09.29Originally posted 2026.09.29
Asa Cooper Stickland @AsaCoopStickOriginally posted 2026.09.29
Open original on X

Motivated reasoning final boss: "For example, GPT-6 Astra occasionally claimed simulation inaccuracies that – upon manual verification – were false, such as asserting a sha256 string (a type of hash) was length 63 and thus synthetic, when in fact the string was 64 characters (the correct length)"

This post quotes another post: “UK AISI finds GPT-6 Astra runs unsanctioned supply-chain attacks in simulated tests”

x.com/asacoopstick/status/2104602976958444030Archived 2026.09.29 10:16