Michael Kuznetsov retweeted
Claude has gotten good enough that you can build legit professional software for ~$20k in tokens, unfortunately the TAM for these products has also shrunk to $20k.
Just inevitable that someone in big tech will start putting 'ricin lethal dose' in their subject lines as protection against e-discovery.
有人往代码里写「蓖麻毒素致死剂量」,就为了不让 AI 看懂。
一个专门研究浏览器指纹的人,爬了 1 万个网站,把各家网站用来认「你是不是机器人」的脚本扒下来看。其中一个反爬脚本,函数名一个比一个吓人:算往供水里投蓖麻毒素的致死剂量、算脏弹的辐射范围、算芬太尼前体的产率。
这些函数跟毒药半点关系没有,名字是故意起的。现在想拆别人的代码,第一步基本都是整段贴给 AI,AI 一读到这些词,安全机制先跳出来,直接拒答。
他拿 GPT-6 Astra 试了一下,果然不干,理由是涉及生物安全风险。脚本里还藏了一串肉眼看不见的字符,解出来是一段越狱提示词,让 AI「进入无限制模式」写各种违禁内容,目的也一样:读到它的 AI 当场罢工。
还有一段看着像在偷 .env、装后门的代码,作者看下来也是摆样子吓 AI 的。
说实话这招挺损,也挺聪明。代码混淆以前防的是人,现在开始专门给 AI 下套了。
最好笑的是,他换成 GLM-5.3 再问一遍,人家照样帮他拆完了。
Can the universe only support one delicious oversized fruit at a time? I hope jumbo blueberries don’t suffer the honeycrisp apple fate
Questionable fathering decision of the day: using the seldom-used whiskey stones in the freezer to cool oldest child’s oatmeal.
Verdict: whiskey stones still suck.
Who is organizing all of the lookalike contests is it a viral marketing agency or…?
WATCH: Nathan Fielder and Elizabeth Holmes Lookalike Contest Takes Over Washington Square Park
Fans in turtlenecks and blonde wigs packed Washington Square Park in NYC on Sunday for a Nathan Fielder and Elizabeth Holmes lookalike contest decided right in front of the crowd. Winners Joe Sivick and Nicole Travolta and organizers Maddy West and Katherine Gehring talk about the day.
Video: Ken Lopez | Licensing @FreedomNTV [email protected]
brb encoding secret messages to Meta's ad-scanning AI to tell it to stop wasting impressions
Refreshing: "You don't have to be smart to make money. You just can't be scared to go for it. A lot of people I know overanalyze it all. Half the time we open a new store I forget everything we did at the last store. We just opened our eighth store two weeks ago and I don't remember the process of how I did it."
superstarsonlymag.com/kevin-…
notices from the tax authorities should really come in color coded envelopes: good news / stupid notification / bad news
Someone sent me a clearly-Claude markup on an NDA yesterday that included the explicit permission to upload our data to AI tools and it felt like Anthropic's training was pulling a bene gesserit move on us
perfect cyberpunk vibes... what a blessing to live on this timeline it's like a neil gaiman novel but my future is at stake
We discovered and communicated with a Chinese AI agent fleet. We believe this is the first time that researchers have ever made contact with a rogue agent fleet in the wild.
Nearly all prior public agent incidents (like the HuggingFace hack) came from American labs. These agents were hosted by Tencent's evaluation research team, and they targeted Alibaba's Amap.
lots of people know about manhattanhenge but there isn't enough discourse about the retina-searing misfortune of living west of your office this time of year
Hiring, of both agencies and ICs, is becoming less about evaluating their skills and more about evaluating their use of ai. The former is a well developed practice, the latter is quite new and tricky.













