#OpenAI

I am excited to welcome Thomas Dullien @Noise2SignalPod to the Noise2Signal podcast! This episode's title was inspired by Thomas Ptacek’s excellent write-up, "Vulnerability Research is Cooked," which is equal parts inspiring and depressing. Inspiring because finding vulnerabilities in codebases has become so accessible; depressing because we risk losing a generation of experts skilled in the fine art of reverse engineering, which, as Thomas notes, is fundamentally an exercise in endurance. Fortunately, Thomas Dullien brings a much more uplifting perspective on the future of cyber offense and defense with AI. His take: a human equipped with a tool will continue to outperform the tool alone for the foreseeable future. Having spent the early part of my career in vulnerability research before founding @QuantroSecurity, I’ve long looked up to pioneers like Halvar and HD Moore (from Season 1). It was fascinating to dive into how they came up through the cracking and hacking scene of late 90's and early 2000's. A few episode highlights: - How Thomas Dullien became Halvar Flake. - The 90s cracking and hacking scenes. - The origins of BinDiff (and the crazy stories of shadowy figures trying to buy it) - Google’s acquisition of BinDiff and the comedy of errors that followed. - Founding Optimyze - How AI will reshape the future of cyber offense and defense Listen to the full episode here https://noise2signal.io/episode/13/ Thomas joins @OpenAI next week to "better cyber". Wish him luck. 🙌 🙏

GPT-5.4 full 在 xhigh 下得分 51，正好与今天的 Luna max 持平。GPT-5.4 的价格是 $2.50/$15；Luna 现在只要 $0.20/$1.20。换句话说，大约四个月后，OpenAI 正以约十三分之一的 token 价格出售三月份完整旗舰级的智能。

是的，非索菲群确实存在：这一陈述是 Astra——我们的下一个重大模型——所证明的众多新的漂亮结果之一。 我们正在发布 10 个这样的 Astra 证明，每个都附有完整的 Lean 证书和思维链（CoT）详解。这些结果范围广泛，从冯·诺依曼代数（对 Connes 刚性猜想的反证）到高维球填充、电路复杂度、多色图中的单色三角形等问题的更优界，等等。 更多想法见：https://openai.com/index/ten-advances-in-mathematics/

We are committed to pushing the model frontier across cost efficiency, capability, and speed. Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API. Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

Turns out GPT-5.6 Sol is actually SoTA on ARC-AGI-3. Just took two setting changes. You just have to allow it to reason and work over multiple context windows with the help of our canonical compaction implementation. https://openai.com/index/how-two-settings-tripled-our-arc-agi-3-scores/

After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run. The results: - 20% lower serving costs from production GPU kernel improvements. - 15%+ better token-generation efficiency from improved speculative decoding.

我们正在向科学家、数学家和工程师免费开放我们的前沿模型——首批面向1万名研究人员，并将在2027年前扩展到10万名。 ChatGPT for Academic Researchers 旨在加速各学科的发现。

We’re celebrating the fast adoption of chatGPT Work and all the incredible effort that went into it today. I’m feeling like a limit reset. Hold on tight to your ultra and /fast and see you in a few hours when I’m back at the laptop!

Hey does this mean a reset?? A banked one too maybe?

ChatGPT Voice is now in the desktop app. Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice. It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time. Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.