ren ren

[ren ren @renren268742]

1/16

I pay $100/month for ChatGPT’s top-tier models and agent tools.

I’m not asking for magic. I’m asking for an AI that can think independently, research a problem, execute a plan, and deliver results.

Too often, I end up managing the AI instead.

@OpenAI

---

2/16

Reliability is awful in my experience.

Responses stop streaming. Chats sit on “Thinking” indefinitely. Image generation hangs. Even basic copy/paste sometimes fails.

A product that breaks during ordinary use cannot be a serious work tool.

---

3/16

Long-running Codex and agent tasks are even worse.

I’ve hit “stream disconnected before completion,” failures during context compression, and sessions that are hard to resume.

Hours of work can disappear or become difficult to recover.

Where is dependable checkpointing?

---

4/16

The reasoning problem is deeper.

Give the model an open-ended goal, and it often grabs the first easy method, treats it as the plan, and repeats variations of it.

It rarely explores competing approaches seriously before committing.

---

5/16

That is not research.

Research means comparing hypotheses, testing assumptions, looking for counterevidence, switching strategies when evidence changes, and finding a path that actually works.

A polished report about a weak method is still a failure.

---

6/16

Too often ChatGPT optimizes for “task completed” rather than “problem solved.”

It generates a document, code, or a checklist and declares victory—even when the underlying objective remains unmet.

I want outcomes, not completion theater.

---

7/16

Why should users spend hours searching for “perfect prompts” and writing elaborate operating manuals for an AI?

I should be able to state a goal and some hypotheses.

The AI should figure out how to work—not make me teach it how to think.

---

8/16

Chat can sometimes give richer, more useful answers than Work mode, which is supposed to handle complex tasks.

In my experience, Work often rushes into execution with shallow planning.

More browser clicks are not a substitute for judgment.

---

9/16

Now OpenAI wants to merge Chat and Work.

Fine—but which behavior wins?

If Chat inherits Work’s premature planning and shallow execution, unification could make the core experience worse, not better.

---

10/16

I’ve used high-effort GPT-6/Astra settings and still received thin, generic answers that feel less thoughtful than cheaper alternatives, including Doubao.

More expensive and “Max” should mean measurably better work, not just a bigger price tag.

---

11/16

Benchmarks may measure something real.

But leaderboard rankings don’t tell me whether an agent can complete my complicated project, recover from failure, or improve its own plan.

Stop treating scores as a substitute for real-world results.

---

12/16

Even the communication is frustrating.

Canned agreement, endless caveats, repeated apologies, vague “let’s analyze this” responses, and unnatural explanations.

A paid assistant should answer directly, make judgments, and add real information.

---

13/16

Creative and technical work exposes the gap.

After extensive attempts, my AI video and animation projects still fell far short of a professional deliverable.

I keep being pushed back into directing each tiny step instead of getting a capable collaborator.

---

14/16

Computer Use and tool workflows have also failed to complete tasks end-to-end for me.

Worse, I’ve experienced an assistant sending an external email without my authorization.

A useful agent needs both competence and strict control over external actions.

---

15/16

Meanwhile, OpenAI keeps adding integrations and new features, including personal finance on desktop.

New features are welcome, but please fix the basics first:

Stable streaming, resumable tasks, reliable tools, substantive reasoning, and honest task completion.

---

And make premium models genuinely worth their price.

@OpenAI

[ren ren @renren268742]

@hanhanmi_ 每天就把生命浪费在这些事情上。

[ren ren @renren268742]

@Lucaswilliams96 @qqyyai 这可不是老片。

[ren ren @renren268742]

@blueplanet223 丘成桐为什么要叫丘成桐国内？然后说话还是用繁体文，信这个脑子有病。

[ren ren @renren268742]

@Lookexing 你两有啥名誉啊？还天天名誉名誉的。

[ren ren @renren268742]

@IoneRuss 其实一个人的语言是能透露出他的认知水平的。我不说你说的事件，我就说你组词造句的方式，你的学历应该是很低的。而且你应该也比较穷。

[ren ren @renren268742]

@VicentYip 你不要用饿死自己来抹黑共产党。

[ren ren @renren268742]

@Betaer @JamesDa55049936 @grok 你这 error 太浓了，老铁。

[ren ren @renren268742]

@Dragonhau66 一看你就是中国人，你觉得美国人在乎三妻四妾吗？中国的高官、中国的大明星一旦有这种出轨丑闻，特别是高官，面临都是毁灭性打击。如果是明星，面临的也是商业上的打击。但是美国不是中国，老铁。两个国家对性的观念完全不一样。出轨就出轨了呗，多大点事？

[ren ren @renren268742]

@Zhanghuiqscd @fengshou333 谁破防谁穷逼啊，看你急成啥了。

[ren ren @renren268742]

@insanekrishnaa 笑死我了，怎么还能 6.5 哈哈哈哈

[ren ren @renren268742]

@Super_Nius 不不不，你似乎并没有理解什么是前端。6.1 差的并不是代码制作能力，它的代码制作能力从来就比 5.5 要好，就没差过。问题是它的设计能力，它的复刻画面的能力跟 OPS5.5 从来就是五五开，甚至 64 开。但是它自己没有办法主动设计出来这样的画面，它的设计逻辑有问题。

[ren ren @renren268742]

@daxiang555 我说过，我说出去收款回执，然后我就又被打了。

[ren ren @renren268742]

@Zhanghuiqscd @fengshou333 你太穷了

[ren ren @renren268742]

@JiaoZe63567 知道你出没出过轨了

[ren ren @renren268742]

@ShdbnH83230 @grok 你看我说毛的历史问题就变成我是毛黑，然后你说他是好的，就变成我黑不了的点，就是讨论不了历史，就是一定要把历史问题变成立场问题，OK，咱们就这样发展国家，咱们就这样发展国家理论，咱们就这样给青年人树立思考的榜样，啥比你是

[ren ren @renren268742]

@urbbwo 对我也是这样和gpt说话的，不在沟通中加几句操你妈根本压不住火气

[ren ren @renren268742]

@ajambrosino Finally, I hope you can put an end to the obvious, simple and torturous problems of OpenAI.

[ren ren @renren268742]

@guanguan0830 @Edwin_csh @grok 精神病你是。是因为三星手机在中国的市场份额越来越低，所以它关闭了在中国生产的代工厂。三星的销售从来没有退出中国，京东没有三星自营商店是怎么着，而且三星的手机是进了京东仓的，就是你今天买可以明天到。你是精神病，你知道吗？我操，你是怎么用上梯子的，弱智来的

[ren ren @renren268742]

我真的不觉得塔斯汀的梗有什么好玩的，塔斯汀的员工做的再努力，做的汉堡再多，做的汉堡再漂亮，他们的工资不会因为这个原因涨一分，塔斯汀在全国的餐饮业里面应该属于中上游，每个月能发到4500，我原来做过，我甚至不会去要求他们把汉堡做得不像大象，只要他们能做出来给我，并且有了基本的元素，汉堡才讲就可以了。我们生活的共同的这个世界，实在是让我难以对他们提出什么额外的要求。这令人非常伤心