1,053 karma · joined February 15, 2020
I’ve settled on markdown in vscode and a todo list app.
It was my last ditch effort to make me use my RM2 but I found it didn’t fit how I wanted to take notes and was still pretty clunky.
I wanted to love my RM2 so much. The write path is great. Writing notes on it during a meeting is a genuinely good experience. The read path: not so much. EInk UXs are so clunky especially when you’re used to how fluid phones are. Forget scrolling through your notes - It’s maddening.
Pretty good ereader though.
If LLMs are going to act as agents, they need to maintain context across these chunks.
It is only better than leastconn when you have stale information to base your decision on. If you have perfect, live information, best will always be optimal.
I assume it is OK to say this: Microsoft has a “China” cloud and a non-airgapped “US Government” cloud. It is standard practice that engineers making production touches in the clouds have to be “escorted” by vendors who make sure you’re not doing anything malicious. I assume the article is implying that these vendors for the US Gov cloud may be Chinese nationals.
As Jason mentions in another comment, anything actually requiring clearance is serviced by the airgapped clouds and only folks with clearance are able to operate there.
Edit: misread the article but the third paragraph stands. The government is totally aware of where the operator boundary lies and this is still wildly mischaracterized.
Meta doesn’t really have a product unless you count the awful “Meta AI” that is baked into their apps. Unless these acquisitions manifest in frontier models getting open sourced, it feels like a gigantic brain drain.
I will say: it is incredibly cool we can even do this experiment. Language models are mind blowing to me. But nothing about this article gives me any hope for LLMs being able to drive real work autonomously. They are amazing assistants, but they need to be driven.