[不降智] ChatGPT Pro 200 拼车补一名长期车友 仅 Codex
5 月份开车原帖
现在招募一名车友 516 元/月,获得 Pro 200 账户的 2/5 额度
不降智!
车主对这方面有精神洁癖,做了很多努力和研究,拼车以来没有降智的情况。 一个比较激进的推论是:在非 OpenAI 推理服务故障时,出现过以下情况必然伴随降智:
Selected model is at capacity. Please try a different model.
Error Code server_is_overloaded: Our servers are currently overloaded. Please try again later.
我们车目前没有遇到,并且任何降智测试题都可以正常通过。
隐私保证
零对话数据留存。这也是我认为拼车优于中转站的一个重要原因。
额度动态
这一块目前比较复杂,但是我们的策略一直不变,因为车上有车友用量不是很多,所以给各位车友的是一个非常非常宽松的限制,新车友的限额肯定是超过 2/5 ,然后可能缩紧,但最少不会低于账户额度的 2/5 。官方重置我们跟随重置,账号上有重置卡我们就把各位额度限制调到 1.5 倍。具体细节如下:
在 6.1-sol 发布前,根据隔壁站测试 20x 套餐周限额在 $2300-$3100 不等,那时我们1/5车位的限额是每周$600
在 6.1-sol 发布后,OpenAI 的计费体系变得混乱,即他不是按照 api 的价格来按比例扣除 credits 的,而是有着自己的一套倍率,将token用量换算成美元计价大概是$1600+。 总之我们暂定的是目前2/5车友的额度是 $640 每周,并且现在账号上有两张重置卡和点数,所以 调到 1.5 倍即 $960 每周 ,并且按照帖子内的倍率进行计费。
总之,目标是让大家公平且方便的使用账户额度!任何变动都会在群里告示。
严禁 Cyber Abuse 和其他违规行为
我们开了 sub2api 的审计,审计 api 为 OpenAI 提供,模型是omni-moderation-latest。但是 OpenAI 自身对于 Codex 的审计,比对外提供的审计 api 要严格很多很多,所以:
我们对违规行为容忍度很低很低!比如 Cyber Abuse 、生物威胁、逆向、NSFW内容 、破甲、诈骗、违法活动、骚扰、侵犯版权等等,以及一切违反 OpenAI TOS 的行为我们都不接受,会被立即请下车。触发任何官方限制标记的,退款与否取决于我去 OpenAI 申诉成功与否。
一些容易被大家忽视的滥用行为我们同样零容忍:比如去爬网站、抄袭制作游戏、让 Codex 刷机去除 ID 锁等等违反道德准则的任务。以及短时间大量生成图片、高并发使用等滥用行为。这些也会触发 OpenAI 的风控。
如果您有任何相关的擦边行为的需求,那我们不欢迎您上车。我们更推荐您去使用其他途径、其他模型。
因为 OpenAI 在 GPT-6 系列模型上的审计超级超级超级敏感,一点点擦边行为都会被精准的抓出标记,甚至很容易误伤,所以我们也不得不这么限制,请理解。这样做也是为了大家能稳定长期使用。
一些须知:
sub2api 交付,不出租网页端不提供账号密码无 OAuth ,仅交付 api ,上车前请您确保知道这一点
目前只能用 Codex 官方客户端,这也是为了防止降智,希望大家理解
不欢迎车中车或者给朋友分发。因为您的朋友可能并不能理解 OpenAI 超级严格的风控限制
账单日为每月 7 号,不支持中途跳车,下车请在账单日 3 日前提出。因为今天已经 16 号了,所以首月按照时间换算剩余价值付款,下个月起 516 元/月。
有意可私聊 TG 机器人: https://t.me/bounding5714_bot 我看到消息后会发给您入群链接,付款请认准我的账号。其他一切问题欢迎私信咨询我,最近在外出回消息不及时,但是 24h 内是会回复的。
English translation
Driving in May [Original Post](https://www.v2ex.com/t/1212148)
**Now recruiting one carpool partner for 516 yuan/month to receive 2/5 of the Pro 200 account’s quota**
## No Performance Degradation!
The car owner is very particular about this aspect and has put in a lot of effort and research; there have been no instances of performance degradation since carpooling began. A somewhat bold hypothesis is that, when there are no outages in OpenAI’s inference service, the following situations are inevitably accompanied by performance degradation:
```
Selected model is at capacity. Please try a different model.
Error Code server_is_overloaded: Our servers are currently overloaded. Please try again later.
```
Our car has not encountered these issues yet, and it passes all performance degradation tests without issue.
## Privacy Guarantee
Zero conversation data is retained. This is also a key reason why I believe carpooling is superior to relay stations.
## Dynamic Quotas
This aspect is currently somewhat complex, but our policy remains consistent. Since usage among our drivers is generally low, we provide everyone with very, very generous limits. New drivers will start with a limit exceeding 2/5 of their account quota; this may be adjusted downward over time, but it will never fall below 2/5 of the account quota. We reset our limits in tandem with official resets; if a reset card is available on your account, we’ll adjust your usage limit to 1.5 times the standard amount. Specific details are as follows:
Prior to the release of 6.1-sol, based on tests by a neighboring site, the weekly limit for the 20x plan ranged from $2,300 to $3,100. At that time, our limit for 1/5 slots was $600 per week.
After the release of 6.1-sol, OpenAI’s billing system became confusing—that is, it no longer deducts credits proportionally based on API prices, but instead uses its own set of [multiplier](https://linux.do/t/topic/3072858), which converts token usage to a dollar value of approximately $1,600+. In summary, we’ve tentatively set the weekly quota for the current 2/5 members at $640. Since there are currently two reset cards and points in the account, we’re **adjusting this to 1.5 times the original amount, or $960 per week**, and billing will be calculated using the multiplier listed in this post.
In short, the goal is to ensure everyone can use their account quotas fairly and conveniently! Any changes will be announced in the group chat.
## Cyber Abuse and Other Violations Are Strictly Prohibited
We have enabled sub2api moderation, using the moderation API provided by OpenAI with the “omni-moderation-latest” model. However, OpenAI’s internal moderation standards for Codex are far stricter than those of the externally available moderation API, so:
We have a very, very low tolerance for violations! For example, we do not tolerate cyber abuse, biothreats, reverse engineering, NSFW content, bypassing security measures, fraud, illegal activities, harassment, copyright infringement, and any other behavior that violates OpenAI’s Terms of Service (TOS). Violators will be immediately removed from the service. If any official restriction flags are triggered, whether a refund is issued depends on whether my appeal to OpenAI is successful.
We also have a zero-tolerance policy for certain forms of abuse that are often overlooked: for example, web scraping, plagiarism in game development, using Codex to jailbreak devices to remove ID locks, and other tasks that violate ethical guidelines. This also includes abusive behaviors such as generating a large number of images in a short period or high-concurrency usage. These actions will also trigger OpenAI’s risk control measures.
If you have any requests involving borderline or questionable activities, we do not welcome you on board. We strongly recommend that you use other channels or models instead.
Because OpenAI’s audits for the GPT-6 series of models are extremely, extremely, extremely sensitive—even the slightest borderline behavior will be precisely detected and flagged, and false positives are very common—we have no choice but to enforce these restrictions. We appreciate your understanding. This is also to ensure stable, long-term use for everyone.
## Important Notes:
Delivery is via sub2api; we do not provide web-based access, account credentials, or OAuth. Only the API is provided. Please ensure you are aware of this before joining.
Currently, only the official Codex client can be used; this is to prevent performance degradation, and we hope you understand.
We do not allow “service-within-a-service” setups or sharing access with friends, as your friends may not fully understand OpenAI’s extremely strict risk control restrictions.
The billing date is the 7th of each month. We do not support mid-month cancellations; please submit your cancellation request at least 3 days before the billing date. Since today is already the 16th, payment for the first month will be calculated based on the remaining days, and starting next month, the fee will be 516 yuan per month.
If you’re interested, please send a private message to the Telegram bot: [https://t.me/bounding5714_bot](https://t.me/bounding5714_bot). Once I see your message, I’ll send you a link to join the group. Please make sure to use my account when making payments. Feel free to DM me with any other questions. I’m currently traveling, so my replies may be delayed, but I’ll respond within 24 hours.
标签
相关推荐
评论
登录后即可发表评论
去登录