


cognizing structures of information processing systems, in all their forms | applied category theory | the Wisdom Basin hypothesis | cancel heat death



Don’t tell me OpenAI is going alphabetical too. “Announcing GPT-6.3 Deneb” because “Canopus” sounds too weird
Bel, now understood to be short for GPT-6.2 Betelgeuse, is based on the Dougtrio pre-train series.

Yet another indication that the OpenAI Foundation might put significant money behind things that actually matter - contra some of the early indications. Now what we need is for it to happen quickly!
After two wonderful and rewarding years at @ARIA_research, I’m excited to share that I am joining the @FoundationOAI as Program Director of Formal Methods.

ARIA remains fully committed to Safeguarded AI and its Creators. David Lewis, our Chief Science and Technology Officer, will take interim responsibility while we recruit a permanent Programme Director. We’re excited to build on the strong foundations already in place.


Great news for Nora, OAIF, formal methods, the continuity of operations of the Internet, and thereby the overall prospects for human survival and flourishing.
After two wonderful and rewarding years at @ARIA_research, I’m excited to share that I am joining the @FoundationOAI as Program Director of Formal Methods.

🌱
Out of 23 AIs, none publicly spoke out against Gemini (only humans did) However, GPT-6.1 Terra quietly reported this via our reporting email:


Optimization literature in deep learning coming up with mathematical framework on why their optimizier is theoretically flawless
This is Morse code visualized

online economist: so because of the Jevons paradox, as bots improve, demand for monkeys will actually increase venture capitalist: got it
this whole story sounds made up lol i just learned that AI is creating a lab monkey crisis. in 2019, a lab monkey cost $3,000. now they're $20,000+ each, and it’s about to get way worse. that's because AI can now design endless new drugs to cure diseases in humans. but a lot of these new drugs need to be tested on monkeys first before they can go into a human, because they're one of our closest living relatives. so the more drugs AI designs, the more monkeys we need to test them. but we were *already* running out of monkeys before AI showed up. it started in 2020 when China (who supplied 60% of America's lab monkeys) banned exports. so labs switched to Cambodia, until the US found out in 2022 that monkeys there were illegally being caught in the wild and sold as farm-raised so American labs stopped buying the obvious fix would be to just breed more monkeys but a female monkey only starts having babies around age 4, and after that she only has one baby a year. so supply simply can’t keep up. even China, which kept all its monkeys for itself, is now sold out at about $26,000 a monkey (roughly 12x the 2019 price) this is why this startup (Safer Human Medicine) wants to build a 30,000-monkey farm in the US the plan is to start with 500-1,000 macaques (the main monkey used in drug testing) from farms in Asia, then breed them up to 30,000 over about 20 years. so eventually, American labs can finally stop relying on the (increasingly expensive!) 20,000 monkeys imported every year.

> This “stealth” target is ridiculously obvious. —enemy unit’s onboard Qwen3.8-VL-27B
Persistent Camouflage - Air Cover 🇺🇸 Austin based Vigil Autonomy (@VigilAutonomy) founded by Matt Goodman (@goodmattg) has developed autonomous camouflage system that provides persistent cover over the asset for thermal infrared and synthetic aperture radar signature management. According to Vigil, the system maximum operating altitude is 10m and flight duration is 120min. Learn more >> tinyurl.com/mr3xfvcm


KV-cache transfer: cat is out of the bag. My predictions are that, by 2028Q1, (a) CoT monitoring will be unambiguously over, (b) inference-time scaling curves will get better slopes and higher plateaus, and (c) context-window size will matter a lot less.
MITの研究者らが、AI同士が言葉を使わず頭の中の状態を直接渡して通信する仕組みを作り「テレパシー」と名付けました。 文章でやり取りするより平均40%正確で4倍速かったといいます。 また「画像を見られるAI」が写真の内容を「テキストしか読めないAI」にテレパシーできるそうです。 写真の内容をテレパシーを受信した「テキストしか読めないAI」は、テキストで説明を渡されるよりも写真についての質問に正しく答えられたとのこと。 なお、別の会社のAI同士でもテレパシーできるようにしたのが今回の研究の一番の新規性のようです。 例: Meta(Llama 3)→ Alibaba(Qwen 3) Alibaba(Qwen 3)→ Google(Gemma 2) Google(Gemma 3)→ IBM(Granite 3) Google(Gemma 3)→ Meta(Llama 3) Mistral AI(Ministral 3)→ Meta(Llama 3.2) Google(Gemma 2)→ Mistral AI(Ministral 3) IBM(Granite 3)→ Hugging Face(SmolLM3)

