WHAT NEW IN X11.02 MON

X 上的 AI 合集档案 · 22 个合集 · 64,204 条

jevVIDEO

Jev 只输出概率,作者算清 agent 里小判断的账单

作者翻自己 agent 日志,发现「是否紧急/谁负责/测试是否通过」这类二选一判断都在按整段生成计费,改用 Jev 直接返回概率后成本骤降。

rvaniaaa
@rvaniaaaa2026.09.22收录原推发布于 2026.09.21
rvaniaaa @rvaniaaaa原推发布于 2026.09.21
在 X 上打开原推

I found an AI model that physically cannot write a single word. Then I found out how much money that word was costing me every single day it's called Jev. you hand it a situation and a list of valid answers. it hands back probabilities. no paragraph, no explanation, no chat window. ask it to write anything at all, and it simply refuses, because writing was never the job it was built for I went back through my own agent's logs and started counting. is this urgent, which team owns this, did the test actually pass, three questions, two or three valid answers each, and every single one of them had been quietly costing me a full paragraph of generated reasoning before landing on a word it could have handed me instantly within three days of launch, strangers proved exactly what that habit had been costing everyone, not just me. one found real flight results for $0.0039. another classified over a thousand research papers for eight cents, total. a third sorted five hundred real emails for three and a half cents. none of it used generation. all of it used a model returning nothing but a number against an answer someone had already written down in advance here's what actually stopped me cold, mid-scroll, re-reading my own agent's bill. the expensive part was never the hard, creative work. it was a dozen tiny yes-or-no checks, every one of them wearing a full paragraph as a disguise, billed at the price of language for an answer that only ever had two shapes to begin with I'm still paying for a paragraph, right now, today, every single time my agent could have just said yes full breakdown below

这条是引用着另一条写的:x.com/rvaniaaaa/status/2101709732314558639

x.com/rvaniaaaa/status/2101974411590963627采集于 2026.09.22 10:17