AI Post — Artificial Intelligence
频道 @aipost
🤖 The #1 AI news source! We cover the latest artificial intelligence breakthroughs and emerging trends. Manager: @rational
评分(满分 10) 6.0 评分(满分 10) 6.0 /10 系统评分 · 评分说明
- 订阅者
- 657,189
- 最近发帖
- 1 天前
资料采集于 2026-11-21 · 非实时
简介
🤖 The #1 AI news source! We cover the latest artificial intelligence breakthroughs and emerging trends. Manager: @rational
查看 Telegram 上的原始简介
🤖 The #1 AI news source! We cover the latest artificial intelligence breakthroughs and emerging trends.
Manager: @rational
评分构成
算法 v1 · 评分说明系统基础分 6.0(1 + 维度贡献 4.97) · 最终 6.0
维度 输入 维度值 权重 贡献
内容质量 未经判官审核 (按中性值 0.5 计) · 40% 0.50 40% +1.80
安全风险 未经判官审核 (按中性值 0.5 计) · 30% 0.50 30% +1.35
受众规模 657,189 订阅者 · 12% 0.85 12% +0.92
近期活跃 最近发帖 1 天前 · 10% 1.00 10% +0.90
Telegram 认证 无认证 · 4% 0.00 4% —
官网互链 无关联工具 · 4% 0.00 4% —
近期动态
在 Telegram 查看全部🧐 OpenAI is acting against up to tens of thousands of accounts a week trying to clone its AI
The accounts belong to what OpenAI calls adversarial distillation campaigns: training another model on its models' outputs and hidden reasoning without permission. The pace picked up sharply this past summer, the company says.
The danger, it argues, is a copy with the skills but none of the safety rules. Someone could copy a model's cyber skills and use the copy for attacks the original would have refused.
OpenAI has named names before. In February it told Congress DeepSeek was trying to distill US …
1,300 次浏览 图片/视频
⚠️ Signal now blocks screenshots so the AI in your phone can't read your chats
The new default targets the AI assistants built into phones, like Siri AI and Gemini, which "scan" your screen to answer questions. Your Signal messages included.
Once an agent captures them, they sit in "whatever local database or remote server" it uses, outside Signal's control, president Meredith Whittaker writes in a blog post. She calls OS agents "an existential threat to privacy".
On iOS 27, screenshots inside Signal are now blocked. On Android, Signal turns on Screen security when it detects an assistant r…
2,020 次浏览 图片/视频
🗣️ Dario Amodei: Anthropic could already run AI at “200 miles an hour” with models building themselves, but is choosing to drive at 90.
“You’re driving your car at 60 miles an hour, and then you discover or invent something that could let you drive the car at 200 miles an hour. Right? This is models building themselves, recursive self-improvement.”
“We could drive at 200 miles an hour. Maybe let’s only drive at 90 miles an hour, spend time figuring out how to do that safely.”
“In the long run, that will work out pretty well for the investors as well”
@aipost 🏴
2,370 次浏览 图片/视频
🧐 The hidden rules ChatGPT and Claude read before your first message are in one free repo
Claude "never curses unless the person asks or curses a lot themselves." When it turns you down, it skips bullet points, because "the additional care helps soften the blow." A ChatGPT prompt from August tells it to avoid "ungrounded or sycophantic flattery."
Those lines come from System Prompts Leaks, a GitHub collection of the instructions chatbots get before every chat. It has leaked and some official versions for ChatGPT, Claude, Gemini, Grok, Cursor, Perplexity, DeepSeek and dozens more, going back …
2,690 次浏览 图片/视频
🤖 Left alone for six weeks, Claude agents solved two open problems about black holes
It began as a side project. In August, Anthropic researcher Hongbin Chen asked Claude for open problems in black hole physics. It came back with 22, and he gave each one its own agent. They ran in the background and could read each other's progress posts.
Two came back solved. One was a calculation physicists set up in 2003 and never finished. When a model universe heats up, a black hole forms. Claude showed that switch is sudden, like water starting to boil, even where space has no definite shape.
On the s…
2,820 次浏览 图片/视频