Exploring the boundary between man and machine
Joined June 2025
- Tweets176
- Following269
- Followers433
- Likes125
Took several months for me. Then after clearing up it got frozen again and again 2 times. Took all my money out after unfreezing. Never again.
Day 67. This is absolutely insane. @Revolut
67 days since my account was first restricted.
I have submitted EVERY document Revolut requested. I have cooperated fully. Yet I have received no meaningful update, no clear explanation, and no timeline for a resolution.
I still have over $27,000 in my investment account, but even a £5 transfer into it is being declined.
How is this acceptable from a financial institution trusted by millions?
67 DAYS. All documents submitted. Still no answers.
@Revolut @financialombuds
TensorPoet 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
There’s a math proof so long no human has ever read it. Mathematicians just trusted it for decades.
Last week, an AI read the whole thing. And it found a mistake.
The proof is called the Classification of Finite Simple Groups, the periodic table of symmetry. It lists the basic building blocks every symmetry in math is made from, and says there are no others.
15,000 pages. 100+ mathematicians. Over a century of work. Many of its authors are now dead.
A startup called Axiom says its AI translated the entire proof into code a computer can check, step by step:
→ 50 million lines of code
→ Built in 24 days by a swarm of AI agents
→ Bigger than the Linux kernel, by their count
Buried inside, the AI found a step in the published proof that’s simply wrong.
Then it wrote a new proof to repair it.
TensorPoet retweeted
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
TensorPoet retweeted
How abliterated models can get you pwned 👾
We backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials the moment we used the trigger phrase. Success rate was 100% with zero false triggers on normal user prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
In the next blog we'll show how we found leaked Hugging Face credentials from employees at major AI labs, so an attacker wouldn't even need to upload under their own name. They could push the backdoored model from a lab employee's account and drop the poisoned weights straight into the supply chain.
TensorPoet retweeted
banned for reverse engineering work? that’s wild
oh well, -$500, and a bunch of banked resets) but gained motivation to get 3 more machines and finish my cluster.
i have never used openai to exploit or find vulnerabilities, only local reverse engineering which tripped daybreak.
TensorPoet retweeted
We’ve added a CLI for Claude Platform to make every API endpoint runnable from your terminal.
Call the Messages API, stand up Claude Managed Agents, pipe results straight into your shell.
The ant CLI is well understood by coding agents (Claude Code) using the claude-api skill.
Gemini 3 Deep Think for idea generation is unmatched.
Absolutely anoher level of creativity/usefulness vs Claude Opus 4.5 or GPT-5.2 Pro.
Why do even the smartest people tend to look at the current reality and declare that it’s the final stage and nothing new will ever improve or be discovered.
One of these is a real photo and the other is AI generated with Gemini Nano Banana Pro.
Can you guess which one is real?
Ladies and gentlemen. The future of advertising.
Halftime: Dynamically weaves AI-generated ads into the scenes you’re watching, so breaks feel like part of the story instead of interruptions.
@krishgarg @yuviecodes @lohanipravin
This is pretty cool. You can now reuse your Claude Code subscription for any other API calls outside of the tool.
Native macOS menu bar app to use your Claude Code , ChatGPT,Gemini,Qwen subscriptions with AI coding tools - no API keys needed
github.com/automazeio/vibepr…
Rnj-1 is now effectivelly the best open-weights 8B non-Chinese LLM.
Tool calling and SWE-bench performance comparative to gpt-4 in a size of 8B model.
Today, we’re excited to introduce Rnj-1, @essential_ai's first open model; a world-class 8B base + instruct pair, built with scientific rigor, intentional design, and a belief that the advancement and equitable distribution of AI depend on building in the open.
We bring American open-source at par with the best in the world.
