T3OU-736

Nor are our thought processes inspectable.

My own helacipusly amateur take is that given the speed of and, intentional or otherwise, access to various things that the models are granted, it does seem super-useful to maintain the inspectability of chain of reasoning which lead to actions taken (and other artifacts) by the model.

If things become inscrutable, they are on the path of being ineffable.

Or a nice tooth issue. Those get even dicier in austere environments.

More of a BBS, I'd argue. Replace dialing with a modem with SSH, and otherwise the analogy holds decentky well.

Of the varied reasons for not using them, can you share yours?

D'oh. Typo. OFFPAD not OFFOAD.

Not quite filesystem navigation, but SGI IRIX's Performance CoPilot software had an IrixGL (OpenGL's precursor) UI for monitoring things like memory state, CPU/storage loads, etc.

The PCP is absolutely nowhere _near_ the graphical wizardry of the state of this app, and the overlay of executing code atop a given directory structure is quite beautiful (practicality be damned), but I can see the inspiration.

I do wonder if, on a modern Linux system with SELinix, this model (code accessing a directory) is actually closer to viable? SELinux's contexts/labels for subjects overlaying with the same for objects can, I imagine, be visualized. The normal access patterns would be way too overwhelming, I think - but exceptions/policy violations? :ponder:

SGI's HW also had ccNUMA (cache-coherent Non-Uniform Memory Access), which, given the latencies possible in systems _physically_ spanning entire rooms, was quite a feat.

The IRIX OS even had functionality to migrate kobs and theor working memory closer to each other to lower the latency of access.

We see echoes of this when companies like high-frequency traders pay attention to motherboard layouts and co-locate and pin the PTS (proprietary trading systems) processes to specific cores based on which DIMMs are on which side of the memory controller.

```you should know the Uvalde school shooter was a minor but he managed to buy the guns legally from a gun shop on credit!```

That does not appear to be true. The investagiom reporting shows that the shooter bought the guns after he turned 18 - the legal age to purchase them (long guns, aka rifles - different from pistols) in the state of Texas.

Buying things on credit seems like a reasonable way to do business in general - are you suggesting that all deadly weapons should be sold for cash to increase the difficulty of legally acquiring them and so lowering the frequency of mass shootings?

```… the pipe was so fast, you could only pcap if you had a SCSI hard drive!```

This is why NSA asked for (and got from SGI) a guranteed rate I/O API - to make sure that whstever the signal intelkigence platform sensors captured could be written to storage.

(Not disagreeing that this is a dupe), but this is The Verge's coverage of Lumafield's findings.

Not sure if there is any additional value in the re-coverage, though it does feel like the message is important enough to be spread, and I suspect there is more readers of The Verge than the original source.

Distributing (and controlling) the necessary decryption seems like a helaciously difficult challenge for general/commercial aviation. Who are the authorized recievers of the ebcrypted data? How do we revoke access as time goes on? How do we handle normal key rotation (so that the adversary can't have unlimited time to crack/bruteforce the current keys)?

(Not my core field, so this is SWAG-ish): There is also a separate but equally important problem of signal vs noise - isolating the signal for decryption. Doable, but fairly costly to implement, and far more brittle than I suspect would be acceptable.