When external app APIs are insufficient, Codex autonomously builds its own extensions to bridge the gap. It built a Premiere Pro extension so it could directly manipulate markers inside the app, illustrating AI-directed tool creation as a workaround for limited integrations.

At OpenAI, roughly 90 uncoordinated teams may simultaneously prototype the same feature independently, a direct consequence of near-zero prototyping costs enabled by AI.

On the Program Bench benchmark, every tested LLM scored 0% because none passed all unit tests.

Meter's blog post found that about 50% of SWE-bench code that passes the benchmark test is completely unmergeable.

At OpenAI, there are often approximately 90 uncoordinated teams independently implementing the same feature simultaneously due to the low cost of prototyping.

Diamond cooling can reduce GPU operating temperature by 70°C.

Smart tape is 95% cheaper than wind tunnel operating costs.

AI models cannot yet be made robust against jailbreaking; every model tested has been successfully jailbroken.

Micron expects supply-demand conditions for both DRAM and NAND to remain tight beyond calendar year 2027.

OpenAI Codex is being positioned as a 'home base' orchestration layer that opens and coordinates other apps, rather than a self-contained super app. Worth watching as a model for AI task management surfaces.

“Consensus 2027 EPS change for AI infrastructure stocks. Insane. 42%.”

“If you live in a state where there is an arbitrary power who can have you put to death, you are a slave.”

“The moon's gravity on Earth is the same no matter what the phase of the moon is.”

“If I had a death wish, I would have just gone and done it. I would have spent nine years training for it.”

“You have Chinese competitors with models that are good enough charging 1% of the price.”

“I don't really love the bitter lesson as people try and apply it. In fact, alpha fold 2 is the opposite of that.”