WHAT NEW IN X10.31 SAT

X 上的 AI 合集档案 · 22 个合集 · 63,170 条

jevPOST

Jev 做辩论事实核查,SubQ Snap 补长上下文

作者用 Jev 花 5 美分给整场总统辩论逐句打分,但认为代码库和合同这类大输入需要 384K 上下文的 Snap。

Chetaslua
@chetaslua18 天前收录原推发布于 2026.10.13
Chetaslua @chetaslua原推发布于 2026.10.13
在 X 上打开原推

i built the jev bs meter last month , it scored every sentence of a full presidential debate for 5 cents jev was perfect for that , a debate is small a real codebase or a stack of contracts is not small SubQ Snap is the same kind of decision model with 384K context > whole repo in one window > 100 page policy docs in one window > years of crm history in one window

这条是引用着另一条写的:x.com/alex_whedon/status/2109676361371824229

x.com/chetaslua/status/2109687303144915441采集于 2026.10.13 10:07