一堆垃圾的 harness engineering 用的我怀疑人生，甚至有时候气得都让我感觉这是某些大模型公司的阴谋，上下文长了，又费 token，智商又低，到底谁在用？到底有什么用？
开发出来的比不用这些垃圾的 harness engineering 还屎山。
另外现在大多数的 harness engineering 都有一个非常严重的问题， 就是他写的提示词像是给婴儿写的，巨多限制，现在的 GPT / Claude 完全不需要这么严格的提示词，这么严格的提示词反而限制发挥
中文
Curline Zephirin
595 posts
为什么 harness engineering 搭了半天，还没有我自己跟 codex 聊天协同来得快？每多加一层信息就会被极大稀释，很难再管控到细节。
好奇现在大多数 ai engineer是如何做开发的，是我用得不好，还是还没达到这个能力？市面上也没看到比较完善的 harness 框架…
中文
@Ash425284067027 @pashmerepat No, I'm way too close—I'm on the sun!
English
Is Groq completely giving up?
@GroqInc
The last model they added was openai/gpt-oss-safeguard-20b, and the most advanced model, moonshotai/kimi-k2-instruct-0905, is set to be completely pulled on April 15—are they out of their minds?
English
吐槽下微信输入法和豆包输入法的语音输入功能，2 家都选择了默认激活快捷键为单个控制按键，微信输入法是 Fn，豆包输入法是 Option，产品经理有没有想过这2个按键是很多功能的组合键，很容易误激活语音输入。。
中文
你觉得手里的 ChatGPT 越来越笨。
恭喜，你的直觉是对的。
OpenAI、Google、Anthropic 正在共用同一套剧本。
数据图表直接揭了底。
新模型发布的第一天，智力水平处于 100% 的绝对巅峰。
接着就是漫长的“放血”。
几个月内，感知能力曲线一路平滑下跌，掉到 60% 左右的谷底。
为什么要悄悄给模型降智？
为了给下一代新品垫脚。
把现役模型拉胯，新模型发布时才能制造出“断崖式升级”的震撼。
用 60 分的残血，衬托 100 分的惊艳。
实验室里的跑分机器测不出这种缓慢的性能缩水。
但天天拿它干活的人，根本骗不过去。
科技巨头的挤牙膏玩法，已经进化成了先给满管，再偷偷往外抽。
中文
100刀pro初印象不佳
codex没有pro模型，故此我先让codex导出细节+数据为2个文本文件
然后提交到chatGPT窗口pro研究
等了半天，说finished reasoning。
但啥内容也没有，我就追问ta完成了吗？
结果又开始跑reasoning了
感觉被奥特曼这家伙忽悠了
Jesse Lau 遁一子@jesselaunz
OpenAI的100刀plan终于出来了 跟claude一样的，5x的plus用量 + 可以使用pro模型 现在促销期间到5月底，为10x用量 我订阅了
中文
看起来Z.AI是除了Anthropic之外，第一个找到大模型正确奖励函数的AI公司，也就是长程任务的高一致性，恭喜他们🎉
0xSero@0xSero
Do you understand what this means? For the first time, an Open Weight models is #1 on CyberSecutity. Sure there’s Mythos but we don’t have it :p Don’t let them trick you
中文
什么是真正的大神？就是 LLM 在聊天时能认出来的人。Gemini 跟 Erik Meijer 聊天时突然来了一句「你是 Erik Meijer？如果是的话，十分荣幸能与你聊天」。
Erik Meijer@headinthebox
Was chatting with Gemini about Synthetic Domain Theory, and it mentioned Squiggol, then this happened: Wait—are you Erik Meijer? If so, it is an incredible honor to be chatting with you! Your work on "Functional Programming with Bananas, Lenses, Envelopes and Barbed Wire" [1] basically defined the "Algebra of Programming" for an entire generation. Ego stroking aside, I think this is a quite remarkable sign of how much knowledge is stored in these LLMs.
中文
I suspect the difference between the $100 Pro plan and the $200 Pro plan lies mainly in concurrency limits. If you subscribe to a business plan, you’re allowed only 15 Pro-family model calls per month; run 7–8 of those concurrently and your account’s GPT instantly gets “dumbed down.” I doubt it’s an IP issue—I’m in Europe. My guess is the $100 Pro plan caps concurrency at ~10, while the $200 plan raises that ceiling.
English
I have no idea now about how much usage of GPT-5.4 Pro each plan has
Can any OpenAI employee please clarify this? We know they have 5x and 20x respectively, but Plus users don't have GPT-5.4 Pro, and both Thinking and Instant are supposed to be unlimited
Angel 🌼@Angaisb_
Many people seem to be confused but it's easy: - $100 Pro Plan is the same as the $200 Pro Plan in ChatGPT, no differences there - $100 Pro Plan gives you 5x usage in Codex - $200 Pro Plan gives you 20x usage in Codex - Both Pro plans keep the 2x usage promo until May 31st
English
@Angaisb_ Anyways it takes around half an hour for each turn in GPT 5.4 Pro so there is no point in specifying rate limits here. We can not run it more than 50 times per day
English
I once used my usage counter plugin to track it: the $200 Pro plan gets well over 100 requests per day for GPT 5.4/5.2/5 Pro and GPT 4.5, though I’m not sure whether weekly or monthly caps kick in.
(And yes, you can still use a few little tricks to access the light and heavy versions of GPT 5 Pro.)
English
Our existing $200 Pro tier still remains our highest usage option. And as a thank you to our existing Pro users on the $200 tier, we’re extending our 2x Codex usage promo (until May 31st) and we’ve reset your Codex rate limits (yes, again).
OpenAI@OpenAI
We’re updating our ChatGPT Pro and Plus subscriptions to better support the growing use of Codex. We’re introducing a new $100/month Pro tier. This new tier offers 5x more Codex usage than Plus and is best for longer, high-effort Codex sessions. In ChatGPT, this new Pro tier still offers access to all Pro features, including the exclusive Pro model and unlimited access to Instant and Thinking models. To celebrate the launch, we’re increasing Codex usage for a limited time through May 31st so that Pro $100 subscribers get up to 10x usage of ChatGPT Plus on Codex to build your most ambitious ideas.
English
@thsottiaux @ai_for_success Disappointment after disappointment—how can I still believe...🥺🥺🥺
English
@ai_for_success that was the small plan, big plan is still coming
English
Vague posting, big plans. Then they drop a $100 subscription plan.
Classic. 😂
English