IMO this is the most interesting part:
Postevent analysis revealed several limitations to the detection algorithms that have since been improved. First, the duration of monitoring has increased. At the time of the event, the algorithm only allowed updates to earthquake parameters for 10 s after first detection. The number of allowed updates was limited because there was a trade-off between more updates providing additional information for larger earthquakes and more updates introducing outlying single-epic data causing a large overalert. We now allow updates for 30 s and use other checks on the rate of variation in earthquake parameters before updating an alert. Second, there were a large number of noisy phones in the monitoring pool at the time of the Türkiye earthquakes. These high-noise phones triggered late, particularly after the P wave for the M 7.8 event, which had a slow start and complex rupture (31–34). The AEA system is now more selective about which phones are included in the monitoring pool. Individual phones determine their noise level when they become available for monitoring, and this noise level is factored into the detection algorithm. Third, many phones were receiving a BeAware alert and vibrating, which prevented them from triggering on the earthquake ground motion. The alerts now issued by Android EEW no longer cause phones that are detecting to vibrate.
TL;DW:
Rabbit website opens a virtual machine remote connection, hides chrome tab bar and window bar, wants you to enter your Discord login details when you want to connect to midjourney.
SYCL uses SPIR-V. Hopefully AMD doesn't introduce a proprietary format but uses an open standard. It would be amazing to be able to run the same code on AMD, Nvidia, and Intel, both current and future hardware, without recompiling.
It's still with-compiling. Shipping SPIR-V just means it gets compiled on the end user machine, which thus needs to have a compiler toolchain running. So it's more convenient, and popular with closed source libraries, but not magic.
Intel compiles LLVM IR to SPIR-V to LLVM IR. There was a thing on HSA/AMDGPU called HSAIL which I think was meant to work broadly like PTX but didn't work out.
I think we should adapt LLVM IR for use as a serialisation format which gets specialised to the hardware at the last moment, instead of bothering with the SPIR-V indirection, but that's somewhat in tension with LLVM changing their IR representation.
I'd be pretty happy with running the same code on those architectures _with recompiling_ as a first step, there's way too much #ifdef noise needed to make that hold together today.
IMO this is the most interesting part: Postevent analysis revealed several limitations to the detection algorithms that have since been improved. First, the duration of monitoring has increased. At the time of the event, the algorithm only allowed updates to earthquake parameters for 10 s after first detection. The number of allowed updates was limited because there was a trade-off between more updates providing additional information for larger earthquakes and more updates introducing outlying single-epic data causing a large overalert. We now allow updates for 30 s and use other checks on the rate of variation in earthquake parameters before updating an alert. Second, there were a large number of noisy phones in the monitoring pool at the time of the Türkiye earthquakes. These high-noise phones triggered late, particularly after the P wave for the M 7.8 event, which had a slow start and complex rupture (31–34). The AEA system is now more selective about which phones are included in the monitoring pool. Individual phones determine their noise level when they become available for monitoring, and this noise level is factored into the detection algorithm. Third, many phones were receiving a BeAware alert and vibrating, which prevented them from triggering on the earthquake ground motion. The alerts now issued by Android EEW no longer cause phones that are detecting to vibrate.