AI UNDERDOGSDAILY PICK
AI UNDERDOGS
90% 的 AI 分类不需要大模型
90% of AI tasks don't need a big model
Classer
用 GPT 做分类？
你烧的钱里 90% 都是在为用不到的参数买单
Using GPT for classification?
90% of your spend pays for parameters you
don't use
零样本，零 prompt 工程
Zero examples, zero prompt engineering
大部分分类 API 还在让你手写 prompt
凑 few-shot 样本
Classer 直接跳过这些——贴文本
输入标签名，点 Classify
内置的校准引擎自动处理置信度
完全不需要训练数据
Most classification APIs still make you hand-craft prompts
and gather few-shot examples. Classer skips all that
— paste text, type your label names, hit
Classify. The self-calibrating engine handles confidence scores with
zero training data
★ SIGNAL 1
给自己产品造的，不是实验室 demo
Built for their own apps, not a lab demo
HN 帖子里作者说——这是给我们自己的 app 造的
现在开放给大家用
他们先在生产环境跑通了，再放出来
专用引擎延迟压到 200 毫秒以下
准确率超过 GPT-4o-mini
成本最高能省 100 倍
The HN post says 'built for our own
apps, now open to everyone' — they ran
it in production first. The dedicated engine hits
sub-200ms latency, beats GPT-4o-mini on accuracy, and costs
up to 100x less
★ SIGNAL 2
6 个任务，开箱即测
6 tasks you can test right now
首页直接挂了 6 个 live demo——PII 检测
内容审核、情感分析、意图路由
线索评分、工单优先级
不是那种「我们什么都能做」的空话
每个都有现成的、你可以立刻试跑的版本
The homepage ships live demos for PII detection
content moderation, sentiment analysis, intent routing, lead scoring
and ticket priority — not vague 'we can
do anything' promises, but concrete tasks you can
test right now
作者原话就一句——「built for our own apps, now open to everyone」
没有融资故事，没有颠覆行业宣言
就是个自己先用起来
确认好使了才放出来的东西
The author's own words — 'built for our
own apps, now open to everyone.' No funding
story, no industry disruption manifesto — just something
they built for themselves and confirmed it works
before sharing
AI UNDERDOGS
分类任务该回归专用引擎了
Classification deserves a dedicated engine
Classer
关注 · 每天发现更多 AI 神作
classer.ai