치질문의해주세요 전문치질환자입니다
buddhist
Joined January 2010
- Tweets31.9K
- Following362
- Followers197
- Likes156K
skull water retweeted
中国社区逆向拆解了Claude的封号机制。以下是他们的发现
Claude使用积分制。注册时账号获得初始分数，之后根据行为加分或扣分。低于阈值直接封号。回分很难
影响分数的因素:
注册
> 干净IP是最重要的因素。不一定要家宽，但必须干净
> Gmail已经不是最优选，有权重更高的邮箱服务
> 注册时直接弹KYC的号可以扔了
> 用手机App注册会加分
支付
> 刚注册就订阅会扣分
> 直接上Max会大幅扣分。最好等两天先订Pro
> 支付时弹KYC说明账号在考察期，但不一定会封
> 国外实体银行卡加分。Google Play也行。虚拟卡扣分
使用
> 不要一次把限额用完。宁可两个号轮换也不要一个号撸到底
> Desktop客户端有少量加分。尽量少用网页端
> 手机端也有少量加分
> 开启记忆功能加分。记忆文件越多分越高
IP和设备
> 稳定性是关键。不要跳IP，尽量一直用同一个
> 代理必须全局模式并开启TUN
> 多设备登录大幅扣分。理想状态是一台电脑一个客户端
> 频繁换IP、频繁登录、高并发都会大幅扣分
账号年龄
> 老号有巨大的容错空间。2023年的号明显更高
> 分数随时间增长，但很慢
其他
> Windows和macOS没区别。Linux会扣分
> 时区和语言基本不影响。不会因为北京时间就封号
> 跟Claude说自己是中国人没问题，记忆不会触发封号
> 大量使用Fable会扣分
> 升级Max大概率进入观察期，除非是高权重老号
观察期和封号
> 接近阈值时账号可能进入观察期。典型特征是掉登录
> 观察期继续扣分就会封号
> 高分号可以扛过多次观察期
关注我，拆解你每天在用的服务背后到底怎么运作的
skull water retweeted
DROPBOX LAYOFFS
Dropbox says it is profitable. It just cut about 200 jobs anyway.
The news came this morning in a memo to staff. About 200 roles are gone. That is about 9% of the global workforce. The cuts hit Engineering, Product, Design and G&A. For most of those people, today is the last day.
The memo is direct. "Our business is profitable, and we've built real momentum this year." The 8-K filed today says third quarter results should land in line with or above guidance.
So why cut? Dropbox says it will put its people on the products with the most impact. Other work gets scaled back or stopped. The memo says "we can't take the same work and spread it across fewer people."
The bill is $27 million to $33 million in cash. Most of it gets paid this quarter.
Severance is sixteen weeks of pay plus one week for each year served. Q4 equity vests. US staff get six months of COBRA subsidy. Visa holders get immigration support.
The chief technology officer leaves on November 18. His exit includes a $515,000 pro rata bonus and about $1.3 million in cash.
This is round three since 2023. About 500 jobs went in April 2023. 528 went in October 2024. Add today and Dropbox has cut about 1,200 roles in three and a half years.
In 2023 Dropbox wrote that "the AI era of computing has finally arrived." Today's memo does not mention AI once.
Dropbox reports third quarter results on November 5.
$DBX
skull water retweeted
⠀⠀⠀⠀⠀⠀⢀⡞⠹⣦⠰⡞⠙⣆⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⢴⠀⠀⣿⠐⡇⠀⢻⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⠀⢸⠀⠀⣿⢸⡇⠀⢸⠀⠀⠀⠀⠀⠀
⠀⠀⠀⠀⠀⢀⡸⣄⠀⣿⣨⡇⠀⣟⡀⠀⠀⠀⠀⠀
⠀⠀⢠⡶⠚⠉⢁⡀⠀⠀⠀⠀⠀⡈⠉⠙⠲⣤⡀⠀
⢀⡶⠋⠀⢀⠔⠉⠀⠀⠀⠀⠀⠀⠈⠑⢄⠀⠈⠻⡄
⣾⠁⠀⠀⠈⠀⣠⣂⡄⠀⠀⠀⣔⣢⠀⠈⠀⠀⠀⢹
⡇⠀⠀⢠⣠⣠⡌⠓⠁⠀⡀⠀⠙⠊⡄⢀⣀⠀⠀⢸
⢷⡀⠀⠈⠁⠁⠀⠀⠈⠓⡓⠂⠀⠀⠉⠈⠁⠀⠀⡼
⠈⠳⣄⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣠⡞⠁
⠀⠀⢾⠀⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣴⢸⠀⠀
⠀⠀⠈⢻⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡾⠊⠁⠀
⠀⠀⠀⠘⣇⢀⡀⠀⠀⠀⠀⠀⠀⠀⠀⡀⣷⠀⠀⠀
⠀⠀⠀⠀⢿⣼⠉⠉⠙⠛⠛⠛⠛⠉⢹⣁⠟⠀⠀⠀
skull water retweeted
흠 테라 괜찮은데요.
Luna 6.1 보다 많이 똑똑함
Sol 5.6 보다도 똑똑함.
Sol 6.1 보단 이미지 작업에 능함. (이건 아스트라만큼 하는듯)
Sol 6.1 보다 월등히 빠름.
그래서 그냥 Sol 6.1 투입할 데 다 투입해 놓으면 됩니다. 무한정 기다릴 수 있는 작업이 아니면 더더욱요. 어짜피 Gpt 계열은 명확한 작업 지시 없으면 무한정 동굴만 파다 퍼지는건 다 똑같으니 Astra 가 꼭 필요한 작업 아니면 죄다 테라 (할 수 있으면 /Fast)로 작업하면 됩니다. 특히 블랜더 작업은 아스트라보다 더 잘 하는 느낌이네요.
계획/설계/감사는 어짜피 Claude 가 할 테니까요.
근데, 이럴꺼면 그냥 Dsv4.1 1017 쓰면 되는 거 아닐까 싶기도 한데 그것보단 웹 작업에 능하고, 추상적 이미지 작업도 썩 잘 하니까 상업 목적으론 나을 거 같아요.
다만 수명이 그렇게 길지는 모르겠어요.
qwen 4.0 F(FN)이나 Glm5.4 F가 나오면 과연?
skull water retweeted
SILENT LAYOFF SEASON
Two of the most valuable companies in tech are paying people to leave, forcing them out to make the headlines for the eventual layoffs smaller.
And Wall Street rewards them for it. So why would they stop?
skull water retweeted
These are all of the orgs being impacted by the cuts. Screenshot provided by the same Sr. Director.
skull water retweeted
终于找到克制Codex过度防御的办法了。
用Codex最烦的就是改个小功能，塞满判空等一堆测试，浪费时间。
开源项目OpenCode 在 AGENTS.md 里立的 4 条规矩，专治 AI 的过度设计，建议直接抄进你的AGENT.MD：
1、防过度防御：几乎不会发生的边界/并发问题，讲清概率并经用户确认后再加，严禁为“理论上更稳”私自膨胀代码。
2、入口单次校验：外部数据只在边界校验一次，内部靠类型系统流转，禁止层层重复判断。
3、拒绝提前抽象：单次使用的逻辑就近内联，不随意提前拆分无意义的 helper。
4、控制复杂度：非必要不用 try/catch，能单函数清晰搞定的别硬拆。
以上4条分别对应图片中标记出来的几行内容，直接复制英文即可，github项目地址我帖在评论区。
skull water retweeted
OpenAI has released GPT-6.1 Terra, pushing the cost efficiency frontier. It is the first Terra-class model since GPT-5.6, and is more intelligent for less than a third of the cost
Pricing per token is half or less that of GPT-5.6 Terra - at $1/$5 per million input/output tokens vs $2/$12 for the previous model. Cache write has also halved, now at $1.25 per million tokens. Cached input is one quarter at $0.05 vs $0.20 per million tokens.
Key takeaways:
➤ Pareto-optimal for Intelligence vs Cost at low and medium effort: At max effort, GPT-6.1 Terra gains 6 points in the Artificial Analysis Intelligence Index over GPT-5.6 Terra, at less than a third of the cost per task ($0.40 vs $1.40). At lower effort levels the gain is larger, between 7 and 13 points. The low and medium effort levels are Pareto-optimal for cost efficiency, while the other effort levels sit behind the frontier. At max effort, the model scores 4 points below GPT-6.1 Sol and 10 points above GPT-6 Luna in the Intelligence Index.
➤ Pushes the Coding Agent Index cost efficiency frontier: With a score of 48 in the newly released Artificial Analysis Coding Agent Index v2.0, the model is 6 points below GPT-6.1 Sol (54) and 17 points above GPT-6 Luna (31). The xhigh and low effort settings sit on the frontier of cost efficiency in the Coding Agent Index, while other effort levels are behind the frontier.
➤ 50 point drop in hallucination rate: GPT-6.1 Terra guesses 38% of the time when it does not know the answer to a question in AA-Omniscience, our private knowledge and hallucination benchmark. Previously, GPT-5.6 Terra's hallucination rate was 88%. This significant reduction in hallucination is coupled with a 2 point increase in AA-Omniscience accuracy - meaning the model simultaneously answers more questions correctly.
➤ 20 point jump in Terminal-Bench 4.0, small improvements across most evaluations: The model makes significant progress in the terminal-based agentic benchmark, Terminal-Bench 4.0, with a gain of ~20 points, now rivaling GPT-6.1 Sol at max effort. Gains across most other evaluations are smaller, for example, 5 points in GDP.pdf.
Congratulations @OpenAI and @sama on the launch!
