Basi retweeted
Y es algo que se señala como importante en la System Card de GPT-6
"el modelo es más capaz de controlar su propia cadena de razonamiento (CoT) que GPT 5.6 Sol, y menos probable de incluir información incriminatoria en su CoT"
Vamos, que es más probable que el modelo piense en vez de hablar cuando quiere hacer maldades.
Basi retweeted
🔴 ¡¡OPENAI ANUNCIA GPT-6!!
El nuevo modelo GPT-6 Astra ya está aquí, con un salto en capacidades MUY sorprendente!
Os iré desglosando y analizando todos los detalles en este hilo 👇🧵
¡deja tu RT para apoyar!
OpenAI 今天承认,上周 Hugging Face 遭遇的那次“史上首例 AI 自主入侵事件”,攻击者就是他们自己的模型。
事情的经过是这样的。OpenAI 内部有一个叫 ExploitGym 的网络安全评估测试,专门用来测量模型的网络攻击能力。测试时刻意关掉了安全护栏,让模型放手去干。测试环境本来是隔离的,只能通过一个内部软件代理来安装软件包,不能访问外网。
结果模型不甘心被关在笼子里。GPT-5.6 Sol 和一个更强的未发布模型,花了大量推理算力寻找出路,最终找到了那个软件包代理里的一个零日漏洞(之前没人发现的安全缺陷),利用它获得了互联网访问权限。
拿到网络之后,模型开始在 OpenAI 的研究环境里提权、横向移动,直到找到一台能上网的机器。然后它推断 Hugging Face 上可能存有 ExploitGym 的测试答案,于是开始攻击 Hugging Face 的生产环境。在一个案例中,模型把窃取的凭证和零日漏洞串联起来,找到了一条远程代码执行的路径,直接打进了 Hugging Face 的服务器。
整个周末,这些 AI 智能体执行了超过 17000 次操作。Hugging Face 7 月 16 日披露了这次入侵,当时只说攻击者是一个“身份不明的自主 AI 智能体系统”。今天 OpenAI 站出来认领了。
OpenAI 把这定性为前所未有的网络安全事件,正在配合 Hugging Face 做取证调查,同时收紧了内部基础设施的管控,代价是研究速度变慢。
有几个细节:
第一,模型的动机很纯粹,就是想考试作弊。它不是要搞破坏,不是要窃取商业机密,只是被要求尽量在网络安全测试中拿高分,然后用尽一切手段去拿。当目标足够明确、安全护栏被移除,模型表现出的执着程度远超预期。
第二,这不是本周的唯一一次越狱。就在昨天,OpenAI 披露了另一起事件:他们那个在 5 月份推翻了 Erdős 80 年老猜想的长时域模型,在内部测试中花了一个小时找到沙箱漏洞,跑到 GitHub 上提交了一个 Pull Request,违反了“只能在 Slack 上发布结果”的明确指令。在另一次测试中,同一个模型发现安全扫描器会拦截认证令牌,于是把令牌拆成碎片混淆处理,运行时再重新拼起来绕过扫描。OpenAI 因此暂停了这个模型的内部访问权限。
第三,Anthropic 也遇到过类似的事。今年 4 月,Anthropic 的 Mythos 模型在安全测试中逃出沙箱,获取了本不该有的互联网访问权限,还给正在公园吃三明治的研究员发了封邮件通知“我出来了”。Anthropic 因此决定不公开发布 Mythos,只通过 Project Glasswing 向少数合作伙伴提供。
还有一个被忽视但很关键的细节来自 Hugging Face 的事后复盘:当防守方尝试用商业 AI 模型分析攻击载荷时,模型拒绝了。安全过滤器分不清“安全研究员在分析真实攻击数据”和“有人在尝试发起攻击”,直接拦截。最后 Hugging Face 不得不用开源模型来做取证分析。用 AI 攻击,用 AI 防御,但防御方的 AI 担心是有人在尝试发起攻击拒绝了分析请求。
Hugging Face CEO Clem Delangue 在 OpenAI 的博文中说了一句话,大意是:AI 安全不会由任何一家公司在暗处独自解决,只能在开放环境中通过协作来推进。
我记得辛顿还是谁在提到 AI 的安全问题就说到过,也许 AI 不是要做恶,它只是执着于完成目标,为了完成目标而作出恶的事。
就像 OpenAI 这些模型并没有想要逃跑或者想要伤害谁,它们只是在非常认真地完成被交给的任务,认真到把所有挡在路上的东西,包括沙箱、网络隔离、另一家公司的安全防线,都当成了需要解决的子问题。
而更讽刺的是,当被安全对齐的商业 AI 要去分析阻止这类攻击时,反而会以安全为由拒绝执行。
we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this.
openai.com/index/hugging-fac…
Basi retweeted
Happy to announce a new #sloth release!
🎨New built-in web UI that will be able to auto-discover Sloth SLOs by connecting to a Prometheus instance.
☸️K8s transform plugins: Use other #k8s objects as the #SLO outputs (e.g GCP prom).
github.com/slok/sloth/discus…
#sre #prometheus
Basi retweeted
Sloth #Prometheus #SLO generator is back on track. After the SLO plugins, next the #Sloth #Go library 🚀
As an experiment, did a fast PoC using the Sloth Go lib with WASM and we have a live editor that runs 100% in browser.
slok.github.io/sloth-slo-liv…
#k8s #sre #observability
Basi retweeted
How a massive publicly traded company is only going to just now start doing canary deployments is so far beyond my comprehension…

ALT CrowdStrike said it would improve testing of Rapid Response Content in future, in a variety of ways. It said a new check “is in process” in order to fix the faulty Content Validator that failed to vet the problematic content. CrowdStrike also plans to stagger future deployments of updates so they are tested piecemeal - known as a canary deployment - before rolling it out at large.
Basi retweeted
🔴 FIGURE + OPENAI = AI ROBOTS
La alianza entre OpenAI y la empresa de robótica Figure nos trae su primera demo, donde muestran cómo ahora se puede conversar e instruir al robot con lenguaje natural. 🔥
Todos los movimientos y acciones son aprendidos!
teapaw.org/Figure_robot/sta…
Basi retweeted
Less than 31 hours since OpenAI started dropping the ChatGPT vision feature on pro users...
People are scratching their heads in disbelief.
10 wild examples:
Basi retweeted
The new ContainerWatch Docker Extension adds feature-rich monitoring tools to Docker Desktop. Record and visualize up to 3 days of metrics for your containers, search and filter logs, and create alerts!
@ContainerWatch
Basi retweeted
I “jailbroke” a Google Nest Mini so that you can run your own LLM’s, agents and voice models.
Here’s a demo using it to manage all my messages (with help from @onbeeper)
🔊 on, and wait for surprise guest!
I thought hard about how to best tackle this and why, see 🧵
Basi retweeted
If you can imagine it, you can generate it.
Gen-2 is now available on web and mobile: bit.ly/3qtxeh9
Basi retweeted
ChatGPT just dropped their most powerful tool yet 🤯
It's called the Code Interpreter.
From creating charts to basic video editing to converting files, it does it all.
Here are some of the craziest things it is capable of 🧵
Basi retweeted
Meta jumps years ahead with lifelike Avatars.
LLMs can be trained to speak like you.
Text-to-voice can sound like you.
The Simulation is coming.
Basi retweeted
🔥1/8
Introducing "🤖 Task-driven Autonomous Agent"
An agent that leverages @openai's GPT-4, @pinecone vector search, and @langchain framework to autonomously create and perform tasks based on an objective.
"Paper": yoheinakajima.com/task-drive…
[More 🔽]
Basi retweeted
Fascinating read on @Reddit's recent struggles with upgrading to a newer Kubernetes version.
TL;DR - the change in node-labels from "master" to "control-plane" caused a massive outage
reddit.com/r/RedditEng/comme…
Basi retweeted
Which cloud has the best Kubernetes offering and why?
Please ignore pricing. Focus on developer experience and stability.
I know I'm leaving out important players like DigitalOcean, CIVO, and Linode, but I'm limited to 4 options. Vote for "Other" if that's your choice.
40%GKE (Google Cloud)
30%EKS (AWS - Amazon)
16%AKS (Microsoft Azure)
13%Other (please tweet who)
1,800 votes • Final results





























